Mosaic_ML_MNIST_evaluation.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "Mosaic_ML_MNIST_evaluation.ipynb",
      "provenance": [],
      "collapsed_sections": [
        "tsbQvFHGUPED",
        "63UIltGwndiR",
        "D0gxa2jS4Goe",
        "MaN4SNzNULWj",
        "Z8WviebQHyWt",
        "GtrGVQUG2EvL",
        "LTgpVHkFnvZk",
        "JbXR1-NKsS1e",
        "BoReBChtcZxC",
        "Cy63ZKBOthEF"
      ],
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/sabeesh90/Advanced_Neural_Network_MosaicML/blob/main/Mosaic_ML_MNIST_evaluation.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "aO-7t1Y7-hV4"
      },
      "source": [
        "# Mosaic ML"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "GZaAFdsdIZ36",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "8eb7f8a7-f420-4b06-919c-6f4041f81344"
      },
      "source": [
        "!pip install mosaicml\n",
        "!pip uninstall pillow\n",
        "!pip install pillow-simd\n",
        "\n",
        "!pip install torchcontrib"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Collecting mosaicml\n",
            "  Downloading mosaicml-0.3.1-py3-none-any.whl (250 kB)\n",
            "\u001b[?25l\r\u001b[K     |█▎                              | 10 kB 31.8 MB/s eta 0:00:01\r\u001b[K     |██▋                             | 20 kB 34.8 MB/s eta 0:00:01\r\u001b[K     |████                            | 30 kB 24.2 MB/s eta 0:00:01\r\u001b[K     |█████▎                          | 40 kB 18.5 MB/s eta 0:00:01\r\u001b[K     |██████▌                         | 51 kB 8.1 MB/s eta 0:00:01\r\u001b[K     |███████▉                        | 61 kB 8.7 MB/s eta 0:00:01\r\u001b[K     |█████████▏                      | 71 kB 7.6 MB/s eta 0:00:01\r\u001b[K     |██████████▌                     | 81 kB 8.5 MB/s eta 0:00:01\r\u001b[K     |███████████▊                    | 92 kB 8.6 MB/s eta 0:00:01\r\u001b[K     |█████████████                   | 102 kB 7.8 MB/s eta 0:00:01\r\u001b[K     |██████████████▍                 | 112 kB 7.8 MB/s eta 0:00:01\r\u001b[K     |███████████████▊                | 122 kB 7.8 MB/s eta 0:00:01\r\u001b[K     |█████████████████               | 133 kB 7.8 MB/s eta 0:00:01\r\u001b[K     |██████████████████▎             | 143 kB 7.8 MB/s eta 0:00:01\r\u001b[K     |███████████████████▋            | 153 kB 7.8 MB/s eta 0:00:01\r\u001b[K     |█████████████████████           | 163 kB 7.8 MB/s eta 0:00:01\r\u001b[K     |██████████████████████▏         | 174 kB 7.8 MB/s eta 0:00:01\r\u001b[K     |███████████████████████▌        | 184 kB 7.8 MB/s eta 0:00:01\r\u001b[K     |████████████████████████▉       | 194 kB 7.8 MB/s eta 0:00:01\r\u001b[K     |██████████████████████████▏     | 204 kB 7.8 MB/s eta 0:00:01\r\u001b[K     |███████████████████████████▌    | 215 kB 7.8 MB/s eta 0:00:01\r\u001b[K     |████████████████████████████▊   | 225 kB 7.8 MB/s eta 0:00:01\r\u001b[K     |██████████████████████████████  | 235 kB 7.8 MB/s eta 0:00:01\r\u001b[K     |███████████████████████████████▍| 245 kB 7.8 MB/s eta 0:00:01\r\u001b[K     |████████████████████████████████| 250 kB 7.8 MB/s \n",
            "\u001b[?25hRequirement already satisfied: torchvision>=0.9.0 in /usr/local/lib/python3.7/dist-packages (from mosaicml) (0.11.1+cu111)\n",
            "Requirement already satisfied: torch>=1.9 in /usr/local/lib/python3.7/dist-packages (from mosaicml) (1.10.0+cu111)\n",
            "Collecting argparse>=1.4.0\n",
            "  Downloading argparse-1.4.0-py2.py3-none-any.whl (23 kB)\n",
            "Collecting torch-optimizer==0.1.0\n",
            "  Downloading torch_optimizer-0.1.0-py3-none-any.whl (72 kB)\n",
            "\u001b[K     |████████████████████████████████| 72 kB 1.3 MB/s \n",
            "\u001b[?25hCollecting pyyaml>=5.4.1\n",
            "  Downloading PyYAML-6.0-cp37-cp37m-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_12_x86_64.manylinux2010_x86_64.whl (596 kB)\n",
            "\u001b[K     |████████████████████████████████| 596 kB 57.9 MB/s \n",
            "\u001b[?25hCollecting torchmetrics>=0.5.1\n",
            "  Downloading torchmetrics-0.6.2-py3-none-any.whl (332 kB)\n",
            "\u001b[K     |████████████████████████████████| 332 kB 59.1 MB/s \n",
            "\u001b[?25hCollecting yahp>=0.0.13\n",
            "  Downloading yahp-0.0.14-py3-none-any.whl (33 kB)\n",
            "Requirement already satisfied: tqdm>=4.62.3 in /usr/local/lib/python3.7/dist-packages (from mosaicml) (4.62.3)\n",
            "Collecting pytorch-ranger>=0.1.1\n",
            "  Downloading pytorch_ranger-0.1.1-py3-none-any.whl (14 kB)\n",
            "Requirement already satisfied: typing-extensions in /usr/local/lib/python3.7/dist-packages (from torch>=1.9->mosaicml) (3.10.0.2)\n",
            "Requirement already satisfied: packaging in /usr/local/lib/python3.7/dist-packages (from torchmetrics>=0.5.1->mosaicml) (21.3)\n",
            "Requirement already satisfied: numpy>=1.17.2 in /usr/local/lib/python3.7/dist-packages (from torchmetrics>=0.5.1->mosaicml) (1.19.5)\n",
            "Requirement already satisfied: pillow!=8.3.0,>=5.3.0 in /usr/local/lib/python3.7/dist-packages (from torchvision>=0.9.0->mosaicml) (7.1.2)\n",
            "Collecting ruamel.yaml>=0.17.10\n",
            "  Downloading ruamel.yaml-0.17.17-py3-none-any.whl (109 kB)\n",
            "\u001b[K     |████████████████████████████████| 109 kB 67.7 MB/s \n",
            "\u001b[?25hCollecting ruamel.yaml.clib>=0.1.2\n",
            "  Downloading ruamel.yaml.clib-0.2.6-cp37-cp37m-manylinux1_x86_64.whl (546 kB)\n",
            "\u001b[K     |████████████████████████████████| 546 kB 53.3 MB/s \n",
            "\u001b[?25hRequirement already satisfied: pyparsing!=3.0.5,>=2.0.2 in /usr/local/lib/python3.7/dist-packages (from packaging->torchmetrics>=0.5.1->mosaicml) (3.0.6)\n",
            "Installing collected packages: ruamel.yaml.clib, ruamel.yaml, pyyaml, pytorch-ranger, yahp, torchmetrics, torch-optimizer, argparse, mosaicml\n",
            "  Attempting uninstall: pyyaml\n",
            "    Found existing installation: PyYAML 3.13\n",
            "    Uninstalling PyYAML-3.13:\n",
            "      Successfully uninstalled PyYAML-3.13\n",
            "Successfully installed argparse-1.4.0 mosaicml-0.3.1 pytorch-ranger-0.1.1 pyyaml-6.0 ruamel.yaml-0.17.17 ruamel.yaml.clib-0.2.6 torch-optimizer-0.1.0 torchmetrics-0.6.2 yahp-0.0.14\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.colab-display-data+json": {
              "pip_warning": {
                "packages": [
                  "argparse"
                ]
              }
            }
          },
          "metadata": {}
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Found existing installation: Pillow 7.1.2\n",
            "Uninstalling Pillow-7.1.2:\n",
            "  Would remove:\n",
            "    /usr/local/lib/python3.7/dist-packages/PIL/*\n",
            "    /usr/local/lib/python3.7/dist-packages/Pillow-7.1.2.dist-info/*\n",
            "    /usr/local/lib/python3.7/dist-packages/Pillow.libs/libfreetype-69f25d5e.so.6.17.1\n",
            "    /usr/local/lib/python3.7/dist-packages/Pillow.libs/libjpeg-ba7bf5af.so.9.4.0\n",
            "    /usr/local/lib/python3.7/dist-packages/Pillow.libs/liblcms2-a6801db4.so.2.0.8\n",
            "    /usr/local/lib/python3.7/dist-packages/Pillow.libs/liblzma-99449165.so.5.2.5\n",
            "    /usr/local/lib/python3.7/dist-packages/Pillow.libs/libopenjp2-b3d7668a.so.2.3.1\n",
            "    /usr/local/lib/python3.7/dist-packages/Pillow.libs/libpng16-bedcb7ea.so.16.37.0\n",
            "    /usr/local/lib/python3.7/dist-packages/Pillow.libs/libtiff-41910f6d.so.5.5.0\n",
            "    /usr/local/lib/python3.7/dist-packages/Pillow.libs/libwebp-122bd20b.so.7.1.0\n",
            "    /usr/local/lib/python3.7/dist-packages/Pillow.libs/libwebpdemux-2db559e5.so.2.0.6\n",
            "    /usr/local/lib/python3.7/dist-packages/Pillow.libs/libwebpmux-ec1d5c76.so.3.0.5\n",
            "    /usr/local/lib/python3.7/dist-packages/Pillow.libs/libz-a147dcb0.so.1.2.3\n",
            "Proceed (y/n)? \u001b[31m  ERROR: Operation cancelled by user\u001b[0m\n",
            "Collecting pillow-simd\n",
            "  Downloading Pillow-SIMD-7.0.0.post3.tar.gz (630 kB)\n",
            "\u001b[K     |████████████████████████████████| 630 kB 6.9 MB/s \n",
            "\u001b[?25hBuilding wheels for collected packages: pillow-simd\n",
            "  Building wheel for pillow-simd (setup.py) ... \u001b[?25l\u001b[?25hdone\n",
            "  Created wheel for pillow-simd: filename=Pillow_SIMD-7.0.0.post3-cp37-cp37m-linux_x86_64.whl size=1206465 sha256=c27a4c5060eaaf629ff808894c50fa76bb9a8ce3ee7c1d318dbf583e0ffb432e\n",
            "  Stored in directory: /root/.cache/pip/wheels/28/97/cd/ea4d241571912e0841b62d72696f5457acf8f5dcd76ba6856e\n",
            "Successfully built pillow-simd\n",
            "Installing collected packages: pillow-simd\n",
            "Successfully installed pillow-simd-7.0.0.post3\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.colab-display-data+json": {
              "pip_warning": {
                "packages": [
                  "PIL"
                ]
              }
            }
          },
          "metadata": {}
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Collecting torchcontrib\n",
            "  Downloading torchcontrib-0.0.2.tar.gz (11 kB)\n",
            "Building wheels for collected packages: torchcontrib\n",
            "  Building wheel for torchcontrib (setup.py) ... \u001b[?25l\u001b[?25hdone\n",
            "  Created wheel for torchcontrib: filename=torchcontrib-0.0.2-py3-none-any.whl size=7532 sha256=a6e18a73263cea993fdddf7fc42afa406c5932c1cb23bb0af469d55361c54c46\n",
            "  Stored in directory: /root/.cache/pip/wheels/91/58/d0/f03811c3e34e1f14031294b5f30d8693689972af874d1225b8\n",
            "Successfully built torchcontrib\n",
            "Installing collected packages: torchcontrib\n",
            "Successfully installed torchcontrib-0.0.2\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "iyvVBtah0ZsX",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 358
        },
        "outputId": "43f45c1d-7f94-4ec3-ed39-907b2b58aa85"
      },
      "source": [
        "from google.colab import drive\n",
        "drive.mount('/content/drive')\n",
        "%cd drive/MyDrive/Colab Notebooks/"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "error",
          "ename": "MessageError",
          "evalue": "ignored",
          "traceback": [
            "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
            "\u001b[0;31mMessageError\u001b[0m                              Traceback (most recent call last)",
            "\u001b[0;32m<ipython-input-2-333b9b643334>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[0;32mfrom\u001b[0m \u001b[0mgoogle\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mcolab\u001b[0m \u001b[0;32mimport\u001b[0m \u001b[0mdrive\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 2\u001b[0;31m \u001b[0mdrive\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mmount\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m'/content/drive'\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      3\u001b[0m \u001b[0mget_ipython\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mmagic\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m'cd drive/MyDrive/Colab Notebooks/'\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.7/dist-packages/google/colab/drive.py\u001b[0m in \u001b[0;36mmount\u001b[0;34m(mountpoint, force_remount, timeout_ms, use_metadata_server)\u001b[0m\n\u001b[1;32m    111\u001b[0m       \u001b[0mtimeout_ms\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mtimeout_ms\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    112\u001b[0m       \u001b[0muse_metadata_server\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0muse_metadata_server\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 113\u001b[0;31m       ephemeral=ephemeral)\n\u001b[0m\u001b[1;32m    114\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    115\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.7/dist-packages/google/colab/drive.py\u001b[0m in \u001b[0;36m_mount\u001b[0;34m(mountpoint, force_remount, timeout_ms, use_metadata_server, ephemeral)\u001b[0m\n\u001b[1;32m    134\u001b[0m   \u001b[0;32mif\u001b[0m \u001b[0mephemeral\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    135\u001b[0m     _message.blocking_request(\n\u001b[0;32m--> 136\u001b[0;31m         'request_auth', request={'authType': 'dfs_ephemeral'}, timeout_sec=None)\n\u001b[0m\u001b[1;32m    137\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    138\u001b[0m   \u001b[0mmountpoint\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0m_os\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mpath\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mexpanduser\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mmountpoint\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.7/dist-packages/google/colab/_message.py\u001b[0m in \u001b[0;36mblocking_request\u001b[0;34m(request_type, request, timeout_sec, parent)\u001b[0m\n\u001b[1;32m    173\u001b[0m   request_id = send_request(\n\u001b[1;32m    174\u001b[0m       request_type, request, parent=parent, expect_reply=True)\n\u001b[0;32m--> 175\u001b[0;31m   \u001b[0;32mreturn\u001b[0m \u001b[0mread_reply_from_input\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mrequest_id\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mtimeout_sec\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
            "\u001b[0;32m/usr/local/lib/python3.7/dist-packages/google/colab/_message.py\u001b[0m in \u001b[0;36mread_reply_from_input\u001b[0;34m(message_id, timeout_sec)\u001b[0m\n\u001b[1;32m    104\u001b[0m         reply.get('colab_msg_id') == message_id):\n\u001b[1;32m    105\u001b[0m       \u001b[0;32mif\u001b[0m \u001b[0;34m'error'\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mreply\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 106\u001b[0;31m         \u001b[0;32mraise\u001b[0m \u001b[0mMessageError\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mreply\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m'error'\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    107\u001b[0m       \u001b[0;32mreturn\u001b[0m \u001b[0mreply\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mget\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m'data'\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;32mNone\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    108\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;31mMessageError\u001b[0m: Error: credential propagation was unsuccessful"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Xs2b5EI_5E8I",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 389
        },
        "outputId": "88d83830-1e1f-48f9-8f63-85a6185bfc93"
      },
      "source": [
        "import torch\n",
        "import torch.nn as nn\n",
        "\n",
        "import torch.nn.functional as F\n",
        "import torch.optim as optim\n",
        "\n",
        "import matplotlib.pyplot as plt\n",
        "\n",
        "import seaborn as sns\n",
        "from torch.optim.lr_scheduler import StepLR\n",
        "from tqdm import tqdm\n",
        "from torchsummary import summary\n",
        "from torchvision import datasets, transforms\n",
        "import time\n",
        "\n",
        "from torchcontrib.optim import SWA\n",
        "\n",
        "# import composer\n",
        "# from composer import DataloaderSpec\n",
        "# from torchvision import datasets, transforms\n",
        "# from composer import Trainer\n",
        "# from composer.algorithms import LabelSmoothing, CutOut, BlurPool, scale_schedule, SWA,SAM, MixUp, SqueezeExcite,ScaleSchedule \n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "error",
          "ename": "ModuleNotFoundError",
          "evalue": "ignored",
          "traceback": [
            "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
            "\u001b[0;31mModuleNotFoundError\u001b[0m                       Traceback (most recent call last)",
            "\u001b[0;32m<ipython-input-1-75642afe6e56>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m     14\u001b[0m \u001b[0;32mimport\u001b[0m \u001b[0mtime\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     15\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 16\u001b[0;31m \u001b[0;32mfrom\u001b[0m \u001b[0mtorchcontrib\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0moptim\u001b[0m \u001b[0;32mimport\u001b[0m \u001b[0mSWA\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     17\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     18\u001b[0m \u001b[0;31m# import composer\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;31mModuleNotFoundError\u001b[0m: No module named 'torchcontrib'",
            "",
            "\u001b[0;31m---------------------------------------------------------------------------\u001b[0;32m\nNOTE: If your import is failing due to a missing package, you can\nmanually install dependencies using either !pip or !apt.\n\nTo view examples of installing some common dependencies, click the\n\"Open Examples\" button below.\n\u001b[0;31m---------------------------------------------------------------------------\u001b[0m\n"
          ],
          "errorDetails": {
            "actions": [
              {
                "action": "open_url",
                "actionText": "Open Examples",
                "url": "/notebooks/snippets/importing_libraries.ipynb"
              }
            ]
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "EhbVgl0OZCqT"
      },
      "source": [
        "# trainer = Trainer(\n",
        "#     model=m,\n",
        "#     train_dataloader_spec=train_dataloader_spec,\n",
        "#     eval_dataloader_spec=eval_dataloader_spec,\n",
        "#     max_epochs=70,\n",
        "#     train_batch_size=128,\n",
        "#     eval_batch_size=128,\n",
        "#     checkpoint_interval_unit=\"ep\",\n",
        "#     checkpoint_folder=\"squeeze_checkpoints\",\n",
        "#     checkpoint_interval=1,\n",
        "#     # device =  'DeviceGPU',\n",
        "#     # device ='cuda',\n",
        "#     algorithms=[\n",
        "#         SqueezeExcite(),\n",
        "#         ScaleSchedule(2)\n",
        "#     ],\n",
        "# )"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "vS1Z3qEl13eZ"
      },
      "source": [
        "#### NORMAL MODEL"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "5U6Fotz712sy"
      },
      "source": [
        "dropout_value = 0.05\n",
        "class BaseNet(nn.Module):\n",
        "    def __init__(self):\n",
        "        super(BaseNet, self).__init__()\n",
        "    #####################################################################################################\n",
        "        # Input Block\n",
        "        self.convblock1 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=1, out_channels=8, kernel_size=(3, 3), padding=0, bias=False), \n",
        "            nn.ReLU(),\n",
        "            nn.BatchNorm2d(8),\n",
        "        ) # input size = 28 output_size = 26, in channels  = 1 out channels = 8\n",
        "\n",
        "        # CONVOLUTION BLOCK 1\n",
        "        self.convblock2 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=8, out_channels=16, kernel_size=(3, 3), padding=0, bias=False),\n",
        "            nn.ReLU(),\n",
        "            nn.BatchNorm2d(16),            \n",
        "        ) # input_size = 26 , output_size = 24, 16 to 16 channels\n",
        "\n",
        "        self.dropout1  = nn.Sequential(nn.Dropout(0.05))\n",
        "    #####################################################################################################\n",
        "        self.pool1 = nn.MaxPool2d(2, 2) # input sizez = 24, output_size = 12\n",
        "    #####################################################################################################\n",
        "\n",
        "        # CONVOLUTION BLOCK 2\n",
        "        self.convblock3 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=16, out_channels=32, kernel_size=(3, 3), padding=0, bias=False),\n",
        "            nn.ReLU(),            \n",
        "            nn.BatchNorm2d(32),\n",
        "            # nn.Dropout(dropout_value)\n",
        "        ) # input size = 12, output_size = 10\n",
        "        self.convblock4 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=32, out_channels=64, kernel_size=(3, 3), padding=0, bias=False),\n",
        "            nn.ReLU(),            \n",
        "            nn.BatchNorm2d(64),\n",
        "            # nn.Dropout(dropout_value)\n",
        "        ) # input size = 10, output_size = 8\n",
        "        self.dropout2  = nn.Sequential(nn.Dropout(0.17))\n",
        "    #####################################################################################################\n",
        "        self.pool2 = nn.MaxPool2d(2, 2) # input sizez = 8, output_size = 4\n",
        "    #####################################################################################################\n",
        "\n",
        "        # self.convblock5 = nn.Sequential(\n",
        "        #     nn.Conv2d(in_channels=64, out_channels=128, kernel_size=(3, 3), padding=1, bias=False),\n",
        "        #     nn.ReLU(),            \n",
        "        #     nn.BatchNorm2d(128),\n",
        "        #     # nn.Dropout(dropout_value)\n",
        "        # ) # input size  = 4,  output_size = 4\n",
        "        # self.dropout3  = nn.Sequential(nn.Dropout(0.17))\n",
        "    #####################################################################################################        \n",
        "        # OUTPUT BLOCK\n",
        "        self.gap = nn.Sequential(\n",
        "            nn.AvgPool2d(kernel_size=4)\n",
        "        ) # output_size = 1 channels  = 16\n",
        "\n",
        "        self.classifier = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=64, out_channels=10, kernel_size=(1, 1), padding=0, bias=False),\n",
        "        ) \n",
        "\n",
        "    def forward(self, x):\n",
        "        # first conv block - teacher\n",
        "        x = self.convblock1(x)\n",
        "        x = self.convblock2(x)\n",
        "        x  = self.dropout1(x)\n",
        "        x = self.pool1(x)\n",
        "\n",
        "        # transition block\n",
        "        x = self.convblock3(x)\n",
        "        x = self.convblock4(x)\n",
        "        x  = self.dropout2(x)\n",
        "        x = self.pool2(x)\n",
        "\n",
        "         # global average pooling layer\n",
        "        x = self.gap(x) \n",
        "        # classification        \n",
        "        x = self.classifier(x)\n",
        "        x = x.view(-1, 10)\n",
        "        return F.log_softmax(x, dim=-1)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "LjFMaq-9BJFL",
        "outputId": "825643f3-b64b-4236-ae71-7b2b613332cc"
      },
      "source": [
        "basenet = BaseNet().to('cuda')\n",
        "summary(basenet, input_size = (1,28,28))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "----------------------------------------------------------------\n",
            "        Layer (type)               Output Shape         Param #\n",
            "================================================================\n",
            "            Conv2d-1            [-1, 8, 26, 26]              72\n",
            "              ReLU-2            [-1, 8, 26, 26]               0\n",
            "       BatchNorm2d-3            [-1, 8, 26, 26]              16\n",
            "            Conv2d-4           [-1, 16, 24, 24]           1,152\n",
            "              ReLU-5           [-1, 16, 24, 24]               0\n",
            "       BatchNorm2d-6           [-1, 16, 24, 24]              32\n",
            "           Dropout-7           [-1, 16, 24, 24]               0\n",
            "         MaxPool2d-8           [-1, 16, 12, 12]               0\n",
            "            Conv2d-9           [-1, 32, 10, 10]           4,608\n",
            "             ReLU-10           [-1, 32, 10, 10]               0\n",
            "      BatchNorm2d-11           [-1, 32, 10, 10]              64\n",
            "           Conv2d-12             [-1, 64, 8, 8]          18,432\n",
            "             ReLU-13             [-1, 64, 8, 8]               0\n",
            "      BatchNorm2d-14             [-1, 64, 8, 8]             128\n",
            "          Dropout-15             [-1, 64, 8, 8]               0\n",
            "        MaxPool2d-16             [-1, 64, 4, 4]               0\n",
            "        AvgPool2d-17             [-1, 64, 1, 1]               0\n",
            "           Conv2d-18             [-1, 10, 1, 1]             640\n",
            "================================================================\n",
            "Total params: 25,144\n",
            "Trainable params: 25,144\n",
            "Non-trainable params: 0\n",
            "----------------------------------------------------------------\n",
            "Input size (MB): 0.00\n",
            "Forward/backward pass size (MB): 0.63\n",
            "Params size (MB): 0.10\n",
            "Estimated Total Size (MB): 0.73\n",
            "----------------------------------------------------------------\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "tsbQvFHGUPED"
      },
      "source": [
        "#### BASEMODEL  - 7K PARAMETERS"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "3AbhnVWQ4wBJ"
      },
      "source": [
        "dropout_value = 0.05\n",
        "class Net(nn.Module):\n",
        "    def __init__(self):\n",
        "        super(Net, self).__init__()\n",
        "    #####################################################################################################\n",
        "        # Input Block\n",
        "        self.convblock1 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=1, out_channels=8, kernel_size=(3, 3), padding=0, bias=False), \n",
        "            nn.ReLU(),\n",
        "            nn.BatchNorm2d(8),\n",
        "        ) # input size = 28 output_size = 26, in channels  = 1 out channels = 16\n",
        "\n",
        "        # CONVOLUTION BLOCK 1\n",
        "        self.convblock2 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=8, out_channels=12, kernel_size=(3, 3), padding=0, bias=False),\n",
        "            nn.ReLU(),\n",
        "            nn.BatchNorm2d(12),            \n",
        "        ) # input_size = 26 , output_size = 24, 16 to 32 channels\n",
        "\n",
        "        self.dropout1  = nn.Sequential(nn.Dropout(0.05))\n",
        "    #####################################################################################################\n",
        "\n",
        "        # TRANSITION BLOCK 1\n",
        "        self.convblock3 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=12, out_channels=8, kernel_size=(1, 1), padding=0, bias=False),\n",
        "        ) # input size  = 24, output_size = 24\n",
        "        self.pool1 = nn.MaxPool2d(2, 2) # input sizez = 24, output_size = 12\n",
        "    #####################################################################################################\n",
        "\n",
        "        # CONVOLUTION BLOCK 2\n",
        "        self.convblock4 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=8, out_channels=12, kernel_size=(3, 3), padding=0, bias=False),\n",
        "            nn.ReLU(),            \n",
        "            nn.BatchNorm2d(12),\n",
        "            # nn.Dropout(dropout_value)\n",
        "        ) # input size = 12, output_size = 10\n",
        "        self.convblock5 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=12, out_channels=16, kernel_size=(3, 3), padding=0, bias=False),\n",
        "            nn.ReLU(),            \n",
        "            nn.BatchNorm2d(16),\n",
        "            # nn.Dropout(dropout_value)\n",
        "        ) # input size = 10, output_size = 8\n",
        "        self.dropout2  = nn.Sequential(nn.Dropout(0.17))\n",
        "    #####################################################################################################\n",
        "\n",
        "        # adding a 1x1 kernel  block here to reduce paramters\n",
        "        self.convblock5a = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=16, out_channels=12, kernel_size=(1, 1), padding=0, bias=False),\n",
        "        ) # input size  = 8, output_size = 8\n",
        "        self.pool2 = nn.MaxPool2d(2, 2) # input sizez = 8, output_size = 4\n",
        "    #####################################################################################################\n",
        "\n",
        "        self.convblock6 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=12, out_channels=15, kernel_size=(3, 3), padding=1, bias=False),\n",
        "            nn.ReLU(),            \n",
        "            nn.BatchNorm2d(15),\n",
        "            # nn.Dropout(dropout_value)\n",
        "        ) # input size  = 4,  output_size = 4\n",
        "        self.dropout3  = nn.Sequential(nn.Dropout(0.17))\n",
        "        self.convblock7 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=15, out_channels=15, kernel_size=(3, 3), padding=1, bias=False),\n",
        "            nn.ReLU(),            \n",
        "            nn.BatchNorm2d(15),\n",
        "            # nn.Dropout(0.03)\n",
        "        ) # input size  = 4, output_size = 4\n",
        "    #####################################################################################################        \n",
        "        # OUTPUT BLOCK\n",
        "        self.gap = nn.Sequential(\n",
        "            nn.AvgPool2d(kernel_size=4)\n",
        "        ) # output_size = 1 channels  = 16\n",
        "\n",
        "        self.convblock8 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=15, out_channels=10, kernel_size=(1, 1), padding=0, bias=False),\n",
        "        ) \n",
        "\n",
        "    def forward(self, x):\n",
        "        # first conv block - teacher\n",
        "        x = self.convblock1(x)\n",
        "        x = self.convblock2(x)\n",
        "        x  = self.dropout1(x)\n",
        "\n",
        "        # transition block\n",
        "        x = self.convblock3(x)\n",
        "        x = self.pool1(x)\n",
        "\n",
        "        # second conv block\n",
        "        x = self.convblock4(x)\n",
        "        x = self.convblock5(x)\n",
        "        x  = self.dropout2(x)\n",
        "\n",
        "        # secocnd transition block\n",
        "        x = self.convblock5a(x)\n",
        "        x = self.pool2(x)        \n",
        "\n",
        "        # third conv block\n",
        "        x = self.convblock6(x)\n",
        "        x = self.dropout3(x)\n",
        "        x = self.convblock7(x)\n",
        "\n",
        "        # global average pooling layer\n",
        "        x = self.gap(x) \n",
        "\n",
        "        # classification\n",
        "        \n",
        "        x = self.convblock8(x)\n",
        "        x = x.view(-1, 10)\n",
        "        return F.log_softmax(x, dim=-1)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "63UIltGwndiR"
      },
      "source": [
        "#### BASELINE MODEL WITH DEPTH WISE  - 7.8K PARAMS"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "BLh2oLhiVSJz"
      },
      "source": [
        "dropout_value = 0.05\n",
        "class Net1(nn.Module):\n",
        "    def __init__(self):\n",
        "        super(Net1, self).__init__()\n",
        "    #####################################################################################################\n",
        "        # Input Block\n",
        "        self.convblock1 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=1, out_channels=10, kernel_size=(3, 3), padding=0, bias=False), \n",
        "            nn.ReLU(),\n",
        "            nn.BatchNorm2d(10),\n",
        "        ) # input size = 28 output_size = 26, in channels  = 1 out channels = 16\n",
        "\n",
        "        # CONVOLUTION BLOCK 1\n",
        "        self.convblock2 = nn.Sequential(\n",
        "            # nn.Conv2d(in_channels=8, out_channels=12, kernel_size=(3, 3), padding=0, bias=False),\n",
        "\n",
        "            # depth wise trials\n",
        "            nn.Conv2d(in_channels=10, out_channels=10, kernel_size=(3, 3),groups = 10, padding=0, bias=False),\n",
        "            nn.Conv2d(in_channels=10, out_channels=12, kernel_size=(1, 1), groups = 1, padding=0, bias=False),\n",
        "\n",
        "            nn.ReLU(),\n",
        "            nn.BatchNorm2d(12),            \n",
        "        ) # input_size = 26 , output_size = 24, 16 to 32 channels\n",
        "\n",
        "        self.dropout1  = nn.Sequential(nn.Dropout(0.05))\n",
        "    #####################################################################################################\n",
        "\n",
        "        # TRANSITION BLOCK 1\n",
        "        self.convblock3 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=12, out_channels=10, kernel_size=(1, 1), padding=0, bias=False),\n",
        "        ) # input size  = 24, output_size = 24\n",
        "        self.pool1 = nn.MaxPool2d(2, 2) # input sizez = 24, output_size = 12\n",
        "    #####################################################################################################\n",
        "\n",
        "        # CONVOLUTION BLOCK 2\n",
        "        self.convblock4 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=10, out_channels=13, kernel_size=(3, 3), padding=0, bias=False), # 14 in case of 13\n",
        "            nn.ReLU(),            \n",
        "            nn.BatchNorm2d(13),\n",
        "            # nn.Dropout(dropout_value)\n",
        "        ) # input size = 12, output_size = 10\n",
        "        self.convblock5 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=13, out_channels=16, kernel_size=(3, 3), padding=0, bias=False), # 14 in case of 13\n",
        "            nn.ReLU(),            \n",
        "            nn.BatchNorm2d(16),\n",
        "            # nn.Dropout(dropout_value)\n",
        "        ) # input size = 10, output_size = 8\n",
        "        self.dropout2  = nn.Sequential(nn.Dropout(0.17))\n",
        "    #####################################################################################################\n",
        "\n",
        "        # adding a 1x1 kernel  block here to reduce paramters\n",
        "        self.convblock5a = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=16, out_channels=12, kernel_size=(1, 1), padding=0, bias=False),\n",
        "        ) # input size  = 8, output_size = 8\n",
        "        self.pool2 = nn.MaxPool2d(2, 2) # input sizez = 8, output_size = 4\n",
        "    #####################################################################################################\n",
        "\n",
        "        self.convblock6 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=12, out_channels=15, kernel_size=(3, 3), padding=1, bias=False),\n",
        "            nn.ReLU(),            \n",
        "            nn.BatchNorm2d(15),\n",
        "            # nn.Dropout(dropout_value)\n",
        "        ) # input size  = 4,  output_size = 4\n",
        "        self.dropout3  = nn.Sequential(nn.Dropout(0.17))\n",
        "        self.convblock7 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=15, out_channels=15, kernel_size=(3, 3), padding=1, bias=False),\n",
        "            nn.ReLU(),            \n",
        "            nn.BatchNorm2d(15),\n",
        "            # nn.Dropout(0.03)\n",
        "        ) # input size  = 4, output_size = 4\n",
        "    #####################################################################################################        \n",
        "        # OUTPUT BLOCK\n",
        "        self.gap = nn.Sequential(\n",
        "            nn.AvgPool2d(kernel_size=4)\n",
        "        ) # output_size = 1 channels  = 16\n",
        "\n",
        "        self.convblock8 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=15, out_channels=10, kernel_size=(1, 1), padding=0, bias=False),\n",
        "        ) \n",
        "\n",
        "    def forward(self, x):\n",
        "        # first conv block - teacher\n",
        "        x = self.convblock1(x)\n",
        "        x = self.convblock2(x)\n",
        "        x  = self.dropout1(x)\n",
        "\n",
        "        # transition block\n",
        "        x = self.convblock3(x)\n",
        "        x = self.pool1(x)\n",
        "\n",
        "        # second conv block\n",
        "        x = self.convblock4(x)\n",
        "        x = self.convblock5(x)\n",
        "        x  = self.dropout2(x)\n",
        "\n",
        "        # secocnd transition block\n",
        "        x = self.convblock5a(x)\n",
        "        x = self.pool2(x)        \n",
        "\n",
        "        # third conv block\n",
        "        x = self.convblock6(x)\n",
        "        x = self.dropout3(x)\n",
        "        x = self.convblock7(x)\n",
        "\n",
        "        # global average pooling layer\n",
        "        x = self.gap(x) \n",
        "\n",
        "        # classification\n",
        "        \n",
        "        x = self.convblock8(x)\n",
        "        x = x.view(-1, 10)\n",
        "        return F.log_softmax(x, dim=-1)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "n = Net1().to('cuda')\n",
        "summary(n, input_size = (1,28,28))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "1D8cc4QOy_A0",
        "outputId": "2c4f5986-01c3-4259-d8e8-0063edb8abe5"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "----------------------------------------------------------------\n",
            "        Layer (type)               Output Shape         Param #\n",
            "================================================================\n",
            "            Conv2d-1           [-1, 10, 26, 26]              90\n",
            "              ReLU-2           [-1, 10, 26, 26]               0\n",
            "       BatchNorm2d-3           [-1, 10, 26, 26]              20\n",
            "            Conv2d-4           [-1, 10, 24, 24]              90\n",
            "            Conv2d-5           [-1, 12, 24, 24]             120\n",
            "              ReLU-6           [-1, 12, 24, 24]               0\n",
            "       BatchNorm2d-7           [-1, 12, 24, 24]              24\n",
            "           Dropout-8           [-1, 12, 24, 24]               0\n",
            "            Conv2d-9           [-1, 10, 24, 24]             120\n",
            "        MaxPool2d-10           [-1, 10, 12, 12]               0\n",
            "           Conv2d-11           [-1, 13, 10, 10]           1,170\n",
            "             ReLU-12           [-1, 13, 10, 10]               0\n",
            "      BatchNorm2d-13           [-1, 13, 10, 10]              26\n",
            "           Conv2d-14             [-1, 16, 8, 8]           1,872\n",
            "             ReLU-15             [-1, 16, 8, 8]               0\n",
            "      BatchNorm2d-16             [-1, 16, 8, 8]              32\n",
            "          Dropout-17             [-1, 16, 8, 8]               0\n",
            "           Conv2d-18             [-1, 12, 8, 8]             192\n",
            "        MaxPool2d-19             [-1, 12, 4, 4]               0\n",
            "           Conv2d-20             [-1, 15, 4, 4]           1,620\n",
            "             ReLU-21             [-1, 15, 4, 4]               0\n",
            "      BatchNorm2d-22             [-1, 15, 4, 4]              30\n",
            "          Dropout-23             [-1, 15, 4, 4]               0\n",
            "           Conv2d-24             [-1, 15, 4, 4]           2,025\n",
            "             ReLU-25             [-1, 15, 4, 4]               0\n",
            "      BatchNorm2d-26             [-1, 15, 4, 4]              30\n",
            "        AvgPool2d-27             [-1, 15, 1, 1]               0\n",
            "           Conv2d-28             [-1, 10, 1, 1]             150\n",
            "================================================================\n",
            "Total params: 7,611\n",
            "Trainable params: 7,611\n",
            "Non-trainable params: 0\n",
            "----------------------------------------------------------------\n",
            "Input size (MB): 0.00\n",
            "Forward/backward pass size (MB): 0.55\n",
            "Params size (MB): 0.03\n",
            "Estimated Total Size (MB): 0.58\n",
            "----------------------------------------------------------------\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "#### depth wise with deeper network"
      ],
      "metadata": {
        "id": "D0gxa2jS4Goe"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "dropout_value = 0.05\n",
        "class Net1(nn.Module):\n",
        "    def __init__(self):\n",
        "        super(Net1, self).__init__()\n",
        "    #####################################################################################################\n",
        "        # Input Block\n",
        "        self.convblock1 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=1, out_channels=10, kernel_size=(3, 3), padding=0, bias=False), \n",
        "            nn.ReLU(),\n",
        "            nn.BatchNorm2d(10),\n",
        "        ) # input size = 28 output_size = 26, in channels  = 1 out channels = 16\n",
        "\n",
        "        # CONVOLUTION BLOCK 1\n",
        "        self.convblock2 = nn.Sequential(\n",
        "            # nn.Conv2d(in_channels=8, out_channels=12, kernel_size=(3, 3), padding=0, bias=False),\n",
        "\n",
        "            # depth wise trials\n",
        "            nn.Conv2d(in_channels=10, out_channels=10, kernel_size=(3, 3),groups = 10, padding=0, bias=False),\n",
        "            nn.Conv2d(in_channels=10, out_channels=12, kernel_size=(1, 1), groups = 1, padding=0, bias=False),\n",
        "\n",
        "            nn.ReLU(),\n",
        "            nn.BatchNorm2d(12),            \n",
        "        ) # input_size = 26 , output_size = 24, 16 to 32 channels\n",
        "\n",
        "        self.dropout1  = nn.Sequential(nn.Dropout(0.05))\n",
        "    #####################################################################################################\n",
        "\n",
        "        # TRANSITION BLOCK 1\n",
        "        self.convblock3 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=12, out_channels=10, kernel_size=(1, 1), padding=0, bias=False),\n",
        "        ) # input size  = 24, output_size = 24\n",
        "        self.pool1 = nn.MaxPool2d(2, 2) # input sizez = 24, output_size = 12\n",
        "    #####################################################################################################\n",
        "\n",
        "        # CONVOLUTION BLOCK 2\n",
        "        self.convblock4 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=10, out_channels=10, kernel_size=(3, 3), groups =10,padding=0, bias=False),\n",
        "            nn.Conv2d(in_channels=10, out_channels=13, kernel_size=(1, 1), padding=0, bias=False), # 14 in case of 13\n",
        "            nn.ReLU(),            \n",
        "            nn.BatchNorm2d(13),\n",
        "            # nn.Dropout(dropout_value)\n",
        "        ) # input size = 12, output_size = 10\n",
        "        self.convblock5 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=13, out_channels=13, kernel_size=(3, 3),groups = 13, padding=0, bias=False),\n",
        "            nn.Conv2d(in_channels=13, out_channels=16, kernel_size=(1,1), padding=0, bias=False), # 14 in case of 13\n",
        "            nn.ReLU(),            \n",
        "            nn.BatchNorm2d(16),\n",
        "            # nn.Dropout(dropout_value)\n",
        "        ) # input size = 10, output_size = 8\n",
        "        self.dropout2  = nn.Sequential(nn.Dropout(0.17))\n",
        "    #####################################################################################################\n",
        "\n",
        "        # adding a 1x1 kernel  block here to reduce paramters\n",
        "        self.convblock5a = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=16, out_channels=12, kernel_size=(1, 1), padding=0, bias=False),\n",
        "        ) # input size  = 8, output_size = 8\n",
        "        self.pool2 = nn.MaxPool2d(2, 2) # input sizez = 8, output_size = 4\n",
        "    #####################################################################################################\n",
        "\n",
        "        self.convblock6 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=12, out_channels=12, kernel_size=(3, 3),groups = 12, padding=1, bias=False),\n",
        "            nn.Conv2d(in_channels=12, out_channels=15, kernel_size=(1, 1), padding=0, bias=False),\n",
        "            nn.ReLU(),            \n",
        "            nn.BatchNorm2d(15),\n",
        "            # nn.Dropout(dropout_value)\n",
        "        ) # input size  = 4,  output_size = 4\n",
        "        self.dropout3  = nn.Sequential(nn.Dropout(0.17))\n",
        "        self.convblock7 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=15, out_channels=15, kernel_size=(3, 3), padding=1, groups = 15, bias=False),\n",
        "            nn.Conv2d(in_channels=15, out_channels=20, kernel_size=(1, 1), padding=0, bias=False),\n",
        "            nn.ReLU(),            \n",
        "            nn.BatchNorm2d(15),\n",
        "            # nn.Dropout(0.03)\n",
        "        ) # input size  = 4, output_size = 4\n",
        "\n",
        "        \n",
        "    #####################################################################################################        \n",
        "        # OUTPUT BLOCK\n",
        "        self.gap = nn.Sequential(\n",
        "            nn.AvgPool2d(kernel_size=4)\n",
        "        ) # output_size = 1 channels  = 16\n",
        "\n",
        "        self.convblock8 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=15, out_channels=10, kernel_size=(1, 1), padding=0, bias=False),\n",
        "        ) \n",
        "\n",
        "    def forward(self, x):\n",
        "        # first conv block - teacher\n",
        "        x = self.convblock1(x)\n",
        "        x = self.convblock2(x)\n",
        "        x  = self.dropout1(x)\n",
        "\n",
        "        # transition block\n",
        "        x = self.convblock3(x)\n",
        "        x = self.pool1(x)\n",
        "\n",
        "        # second conv block\n",
        "        x = self.convblock4(x)\n",
        "        x = self.convblock5(x)\n",
        "        x  = self.dropout2(x)\n",
        "\n",
        "        # secocnd transition block\n",
        "        x = self.convblock5a(x)\n",
        "        x = self.pool2(x)        \n",
        "\n",
        "        # third conv block\n",
        "        x = self.convblock6(x)\n",
        "        x = self.dropout3(x)\n",
        "        x = self.convblock7(x)\n",
        "\n",
        "        # global average pooling layer\n",
        "        x = self.gap(x) \n",
        "\n",
        "        # classification\n",
        "        \n",
        "        x = self.convblock8(x)\n",
        "        x = x.view(-1, 10)\n",
        "        return F.log_softmax(x, dim=-1)\n"
      ],
      "metadata": {
        "id": "nQZ61jpE3wMr"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "m = Net1()\n",
        "summary(m, input_size=(1,28,28))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "HX6YkH0P3VJ-",
        "outputId": "26ecec85-7470-44b4-b57a-bda834eab190"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "----------------------------------------------------------------\n",
            "        Layer (type)               Output Shape         Param #\n",
            "================================================================\n",
            "            Conv2d-1           [-1, 10, 26, 26]              90\n",
            "              ReLU-2           [-1, 10, 26, 26]               0\n",
            "       BatchNorm2d-3           [-1, 10, 26, 26]              20\n",
            "            Conv2d-4           [-1, 10, 24, 24]              90\n",
            "            Conv2d-5           [-1, 12, 24, 24]             120\n",
            "              ReLU-6           [-1, 12, 24, 24]               0\n",
            "       BatchNorm2d-7           [-1, 12, 24, 24]              24\n",
            "           Dropout-8           [-1, 12, 24, 24]               0\n",
            "            Conv2d-9           [-1, 10, 24, 24]             120\n",
            "        MaxPool2d-10           [-1, 10, 12, 12]               0\n",
            "           Conv2d-11           [-1, 10, 10, 10]              90\n",
            "           Conv2d-12           [-1, 13, 10, 10]             130\n",
            "             ReLU-13           [-1, 13, 10, 10]               0\n",
            "      BatchNorm2d-14           [-1, 13, 10, 10]              26\n",
            "           Conv2d-15             [-1, 13, 8, 8]             117\n",
            "           Conv2d-16             [-1, 16, 8, 8]             208\n",
            "             ReLU-17             [-1, 16, 8, 8]               0\n",
            "      BatchNorm2d-18             [-1, 16, 8, 8]              32\n",
            "          Dropout-19             [-1, 16, 8, 8]               0\n",
            "           Conv2d-20             [-1, 12, 8, 8]             192\n",
            "        MaxPool2d-21             [-1, 12, 4, 4]               0\n",
            "           Conv2d-22             [-1, 12, 4, 4]             108\n",
            "           Conv2d-23             [-1, 15, 4, 4]             180\n",
            "             ReLU-24             [-1, 15, 4, 4]               0\n",
            "      BatchNorm2d-25             [-1, 15, 4, 4]              30\n",
            "          Dropout-26             [-1, 15, 4, 4]               0\n",
            "           Conv2d-27             [-1, 15, 4, 4]             135\n",
            "           Conv2d-28             [-1, 15, 4, 4]             225\n",
            "             ReLU-29             [-1, 15, 4, 4]               0\n",
            "      BatchNorm2d-30             [-1, 15, 4, 4]              30\n",
            "        AvgPool2d-31             [-1, 15, 1, 1]               0\n",
            "           Conv2d-32             [-1, 10, 1, 1]             150\n",
            "================================================================\n",
            "Total params: 2,117\n",
            "Trainable params: 2,117\n",
            "Non-trainable params: 0\n",
            "----------------------------------------------------------------\n",
            "Input size (MB): 0.00\n",
            "Forward/backward pass size (MB): 0.56\n",
            "Params size (MB): 0.01\n",
            "Estimated Total Size (MB): 0.57\n",
            "----------------------------------------------------------------\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "MaN4SNzNULWj"
      },
      "source": [
        "#### SMALLER MODEL - 5K"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "SD9ApQb95TR0"
      },
      "source": [
        "class CNet(nn.Module):\n",
        "    def __init__(self):\n",
        "        super(CNet, self).__init__()\n",
        "    #####################################################################################################\n",
        "        # Input Block\n",
        "        self.convblock1 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=1, out_channels=8, kernel_size=(3, 3), padding=0, bias=False), \n",
        "            nn.ReLU(),\n",
        "            nn.BatchNorm2d(8),\n",
        "        ) # input size = 28 output_size = 26, in channels  = 1 out channels = 16\n",
        "\n",
        "        # CONVOLUTION BLOCK 1\n",
        "        self.convblock2 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=8, out_channels=12, kernel_size=(3, 3), padding=0, bias=False),\n",
        "            nn.ReLU(),\n",
        "            nn.BatchNorm2d(12),            \n",
        "        ) # input_size = 26 , output_size = 24, 16 to 32 channels\n",
        "\n",
        "        self.dropout1  = nn.Sequential(nn.Dropout(0.05))\n",
        "    #####################################################################################################\n",
        "\n",
        "        # TRANSITION BLOCK 1\n",
        "        self.convblock3 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=12, out_channels=8, kernel_size=(1, 1), padding=0, bias=False),\n",
        "        ) # input size  = 24, output_size = 24\n",
        "        self.pool1 = nn.MaxPool2d(2, 2) # input sizez = 24, output_size = 12\n",
        "    #####################################################################################################\n",
        "\n",
        "        # CONVOLUTION BLOCK 2\n",
        "        self.convblock4 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=8, out_channels=12, kernel_size=(3, 3), padding=0, bias=False),\n",
        "            nn.ReLU(),            \n",
        "            nn.BatchNorm2d(12),\n",
        "            # nn.Dropout(dropout_value)\n",
        "        ) # input size = 12, output_size = 10\n",
        "        self.convblock5 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=12, out_channels=16, kernel_size=(3, 3), padding=0, bias=False),\n",
        "            nn.ReLU(),            \n",
        "            nn.BatchNorm2d(16),\n",
        "            # nn.Dropout(dropout_value)\n",
        "        ) # input size = 10, output_size = 8\n",
        "        self.dropout2  = nn.Sequential(nn.Dropout(0.17))\n",
        "    #####################################################################################################\n",
        "\n",
        "        # adding a 1x1 kernel  block here to reduce paramters\n",
        "        self.convblock5a = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=16, out_channels=12, kernel_size=(1, 1), padding=0, bias=False),\n",
        "        ) # input size  = 8, output_size = 8\n",
        "        self.pool2 = nn.MaxPool2d(2, 2) # input sizez = 8, output_size = 4\n",
        "    #####################################################################################################\n",
        "\n",
        "        self.convblock6 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=12, out_channels=15, kernel_size=(3, 3), padding=1, bias=False),\n",
        "            nn.ReLU(),            \n",
        "            nn.BatchNorm2d(15),\n",
        "            # nn.Dropout(dropout_value)\n",
        "        ) # input size  = 4,  output_size = 4\n",
        "        self.dropout3  = nn.Sequential(nn.Dropout(0.17))\n",
        "        # self.convblock7 = nn.Sequential(\n",
        "        #     nn.Conv2d(in_channels=15, out_channels=15, kernel_size=(3, 3), padding=1, bias=False),\n",
        "        #     nn.ReLU(),            \n",
        "        #     nn.BatchNorm2d(15),\n",
        "        #     # nn.Dropout(0.03)\n",
        "        # ) # input size  = 4, output_size = 4\n",
        "    #####################################################################################################        \n",
        "        # OUTPUT BLOCK\n",
        "        self.gap = nn.Sequential(\n",
        "            nn.AvgPool2d(kernel_size=4)\n",
        "        ) # output_size = 1 channels  = 16\n",
        "\n",
        "        self.convblock8 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=15, out_channels=10, kernel_size=(1, 1), padding=0, bias=False),\n",
        "        ) \n",
        "\n",
        "    def forward(self, x):\n",
        "        # first conv block-assistant\n",
        "        # x = x.to(\"cuda\")\n",
        "        x = self.convblock1(x)\n",
        "        x = self.convblock2(x)\n",
        "        x  = self.dropout1(x)\n",
        "\n",
        "        # transition block\n",
        "        x = self.convblock3(x)\n",
        "        x = self.pool1(x)\n",
        "\n",
        "        # second conv block\n",
        "        x = self.convblock4(x)\n",
        "        x = self.convblock5(x)\n",
        "        x  = self.dropout2(x)\n",
        "\n",
        "        # secocnd transition block\n",
        "        x = self.convblock5a(x)\n",
        "        x = self.pool2(x)        \n",
        "\n",
        "        # third conv block\n",
        "        x = self.convblock6(x)\n",
        "        x = self.dropout3(x)\n",
        "        # x = self.convblock7(x)\n",
        "\n",
        "        # global average pooling layer\n",
        "        x = self.gap(x) \n",
        "\n",
        "        # classification        \n",
        "        x = self.convblock8(x)\n",
        "        x = x.view(-1, 10)\n",
        "        return F.log_softmax(x, dim=-1)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "tnIi3LU5qRde",
        "outputId": "75806686-ee14-4c62-cd3a-965269d741c8"
      },
      "source": [
        "k5_model = CNet().to(\"cuda\")\n",
        "summary(k5_model, input_size = (1,28,28))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "----------------------------------------------------------------\n",
            "        Layer (type)               Output Shape         Param #\n",
            "================================================================\n",
            "            Conv2d-1            [-1, 8, 26, 26]              72\n",
            "              ReLU-2            [-1, 8, 26, 26]               0\n",
            "       BatchNorm2d-3            [-1, 8, 26, 26]              16\n",
            "            Conv2d-4           [-1, 12, 24, 24]             864\n",
            "              ReLU-5           [-1, 12, 24, 24]               0\n",
            "       BatchNorm2d-6           [-1, 12, 24, 24]              24\n",
            "           Dropout-7           [-1, 12, 24, 24]               0\n",
            "            Conv2d-8            [-1, 8, 24, 24]              96\n",
            "         MaxPool2d-9            [-1, 8, 12, 12]               0\n",
            "           Conv2d-10           [-1, 12, 10, 10]             864\n",
            "             ReLU-11           [-1, 12, 10, 10]               0\n",
            "      BatchNorm2d-12           [-1, 12, 10, 10]              24\n",
            "           Conv2d-13             [-1, 16, 8, 8]           1,728\n",
            "             ReLU-14             [-1, 16, 8, 8]               0\n",
            "      BatchNorm2d-15             [-1, 16, 8, 8]              32\n",
            "          Dropout-16             [-1, 16, 8, 8]               0\n",
            "           Conv2d-17             [-1, 12, 8, 8]             192\n",
            "        MaxPool2d-18             [-1, 12, 4, 4]               0\n",
            "           Conv2d-19             [-1, 15, 4, 4]           1,620\n",
            "             ReLU-20             [-1, 15, 4, 4]               0\n",
            "      BatchNorm2d-21             [-1, 15, 4, 4]              30\n",
            "          Dropout-22             [-1, 15, 4, 4]               0\n",
            "        AvgPool2d-23             [-1, 15, 1, 1]               0\n",
            "           Conv2d-24             [-1, 10, 1, 1]             150\n",
            "================================================================\n",
            "Total params: 5,712\n",
            "Trainable params: 5,712\n",
            "Non-trainable params: 0\n",
            "----------------------------------------------------------------\n",
            "Input size (MB): 0.00\n",
            "Forward/backward pass size (MB): 0.45\n",
            "Params size (MB): 0.02\n",
            "Estimated Total Size (MB): 0.48\n",
            "----------------------------------------------------------------\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "xfrR3B1wqlrd"
      },
      "source": [
        "class CNet1(nn.Module):\n",
        "    def __init__(self):\n",
        "        super(CNet1, self).__init__()\n",
        "    #####################################################################################################\n",
        "        # Input Block\n",
        "        self.convblock1 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=1, out_channels=8, kernel_size=(3, 3), padding=0, bias=False), \n",
        "            nn.ReLU(),\n",
        "            nn.BatchNorm2d(8),\n",
        "        ) # input size = 28 output_size = 26, in channels  = 1 out channels = 16\n",
        "\n",
        "        # CONVOLUTION BLOCK 1\n",
        "        self.convblock2 = nn.Sequential(\n",
        "            # nn.Conv2d(in_channels=8, out_channels=12, kernel_size=(3, 3), padding=0, bias=False),\n",
        "            # depth wise convolution\n",
        "            nn.Conv2d(in_channels=8, out_channels=8,groups = 8, kernel_size=(3, 3), padding=0, bias=False),\n",
        "            nn.Conv2d(in_channels=8, out_channels=12, kernel_size=(1, 1), padding=0, bias=False),\n",
        "            nn.ReLU(),\n",
        "            nn.BatchNorm2d(12),            \n",
        "        ) # input_size = 26 , output_size = 24, 16 to 32 channels\n",
        "\n",
        "        self.dropout1  = nn.Sequential(nn.Dropout(0.05))\n",
        "    #####################################################################################################\n",
        "\n",
        "        # TRANSITION BLOCK 1\n",
        "        self.convblock3 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=12, out_channels=8, kernel_size=(1, 1), padding=0, bias=False),\n",
        "        ) # input size  = 24, output_size = 24\n",
        "        self.pool1 = nn.MaxPool2d(2, 2) # input sizez = 24, output_size = 12\n",
        "    #####################################################################################################\n",
        "\n",
        "        # CONVOLUTION BLOCK 2\n",
        "        self.convblock4 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=8, out_channels=13, kernel_size=(3, 3), padding=0, bias=False),\n",
        "            nn.ReLU(),            \n",
        "            nn.BatchNorm2d(13),\n",
        "            # nn.Dropout(dropout_value)\n",
        "        ) # input size = 12, output_size = 10\n",
        "        self.convblock5 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=13, out_channels=18, kernel_size=(3, 3), padding=0, bias=False),\n",
        "            nn.ReLU(),            \n",
        "            nn.BatchNorm2d(18),\n",
        "            # nn.Dropout(dropout_value)\n",
        "        ) # input size = 10, output_size = 8\n",
        "        self.dropout2  = nn.Sequential(nn.Dropout(0.17))\n",
        "    #####################################################################################################\n",
        "\n",
        "        # adding a 1x1 kernel  block here to reduce paramters\n",
        "        self.convblock5a = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=18, out_channels=12, kernel_size=(1, 1), padding=0, bias=False),\n",
        "        ) # input size  = 8, output_size = 8\n",
        "        self.pool2 = nn.MaxPool2d(2, 2) # input sizez = 8, output_size = 4\n",
        "    #####################################################################################################\n",
        "\n",
        "        self.convblock6 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=12, out_channels=16, kernel_size=(3, 3), padding=1, bias=False),\n",
        "            nn.ReLU(),            \n",
        "            nn.BatchNorm2d(16),\n",
        "            # nn.Dropout(dropout_value)\n",
        "        ) # input size  = 4,  output_size = 4\n",
        "        self.dropout3  = nn.Sequential(nn.Dropout(0.17))\n",
        "        # self.convblock7 = nn.Sequential(\n",
        "        #     nn.Conv2d(in_channels=15, out_channels=15, kernel_size=(3, 3), padding=1, bias=False),\n",
        "        #     nn.ReLU(),            \n",
        "        #     nn.BatchNorm2d(15),\n",
        "        #     # nn.Dropout(0.03)\n",
        "        # ) # input size  = 4, output_size = 4\n",
        "    #####################################################################################################        \n",
        "        # OUTPUT BLOCK\n",
        "        self.gap = nn.Sequential(\n",
        "            nn.AvgPool2d(kernel_size=4)\n",
        "        ) # output_size = 1 channels  = 16\n",
        "\n",
        "        self.convblock8 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=16, out_channels=10, kernel_size=(1, 1), padding=0, bias=False),\n",
        "        ) \n",
        "\n",
        "    def forward(self, x):\n",
        "        # first conv block-assistant\n",
        "        # x = x.to(\"cuda\")\n",
        "        x = self.convblock1(x)\n",
        "        x = self.convblock2(x)\n",
        "        x  = self.dropout1(x)\n",
        "\n",
        "        # transition block\n",
        "        x = self.convblock3(x)\n",
        "        x = self.pool1(x)\n",
        "\n",
        "        # second conv block\n",
        "        x = self.convblock4(x)\n",
        "        x = self.convblock5(x)\n",
        "        x  = self.dropout2(x)\n",
        "\n",
        "        # secocnd transition block\n",
        "        x = self.convblock5a(x)\n",
        "        x = self.pool2(x)        \n",
        "\n",
        "        # third conv block\n",
        "        x = self.convblock6(x)\n",
        "        x = self.dropout3(x)\n",
        "        # x = self.convblock7(x)\n",
        "\n",
        "        # global average pooling layer\n",
        "        x = self.gap(x) \n",
        "\n",
        "        # classification        \n",
        "        x = self.convblock8(x)\n",
        "        x = x.view(-1, 10)\n",
        "        return F.log_softmax(x, dim=-1)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "_C5LowsCrOic",
        "outputId": "5c6db313-5cfd-472c-bb13-ed3840f61f9a"
      },
      "source": [
        "k5_model1 = CNet1().to(\"cuda\")\n",
        "summary(k5_model1, input_size = (1,28,28))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "----------------------------------------------------------------\n",
            "        Layer (type)               Output Shape         Param #\n",
            "================================================================\n",
            "            Conv2d-1            [-1, 8, 26, 26]              72\n",
            "              ReLU-2            [-1, 8, 26, 26]               0\n",
            "       BatchNorm2d-3            [-1, 8, 26, 26]              16\n",
            "            Conv2d-4            [-1, 8, 24, 24]              72\n",
            "            Conv2d-5           [-1, 12, 24, 24]              96\n",
            "              ReLU-6           [-1, 12, 24, 24]               0\n",
            "       BatchNorm2d-7           [-1, 12, 24, 24]              24\n",
            "           Dropout-8           [-1, 12, 24, 24]               0\n",
            "            Conv2d-9            [-1, 8, 24, 24]              96\n",
            "        MaxPool2d-10            [-1, 8, 12, 12]               0\n",
            "           Conv2d-11           [-1, 13, 10, 10]             936\n",
            "             ReLU-12           [-1, 13, 10, 10]               0\n",
            "      BatchNorm2d-13           [-1, 13, 10, 10]              26\n",
            "           Conv2d-14             [-1, 18, 8, 8]           2,106\n",
            "             ReLU-15             [-1, 18, 8, 8]               0\n",
            "      BatchNorm2d-16             [-1, 18, 8, 8]              36\n",
            "          Dropout-17             [-1, 18, 8, 8]               0\n",
            "           Conv2d-18             [-1, 12, 8, 8]             216\n",
            "        MaxPool2d-19             [-1, 12, 4, 4]               0\n",
            "           Conv2d-20             [-1, 16, 4, 4]           1,728\n",
            "             ReLU-21             [-1, 16, 4, 4]               0\n",
            "      BatchNorm2d-22             [-1, 16, 4, 4]              32\n",
            "          Dropout-23             [-1, 16, 4, 4]               0\n",
            "        AvgPool2d-24             [-1, 16, 1, 1]               0\n",
            "           Conv2d-25             [-1, 10, 1, 1]             160\n",
            "================================================================\n",
            "Total params: 5,616\n",
            "Trainable params: 5,616\n",
            "Non-trainable params: 0\n",
            "----------------------------------------------------------------\n",
            "Input size (MB): 0.00\n",
            "Forward/backward pass size (MB): 0.49\n",
            "Params size (MB): 0.02\n",
            "Estimated Total Size (MB): 0.52\n",
            "----------------------------------------------------------------\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Z8WviebQHyWt"
      },
      "source": [
        "#### SMALLEST MODEL"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "CzKuCGZ2FK4a"
      },
      "source": [
        "class SNet(nn.Module):\n",
        "    def __init__(self):\n",
        "        super(SNet, self).__init__()\n",
        "    #####################################################################################################\n",
        "        # Input Block\n",
        "        self.convblock1 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=1, out_channels=8, kernel_size=(3, 3), padding=0, bias=False), \n",
        "            nn.ReLU(),\n",
        "            nn.BatchNorm2d(8),\n",
        "        ) # input size = 28 output_size = 26, in channels  = 1 out channels = 16\n",
        "\n",
        "        # CONVOLUTION BLOCK 1\n",
        "        self.convblock2 = nn.Sequential(\n",
        "            # DEPTH WISE SEPERABLE COVOLUTION\n",
        "            nn.Conv2d(in_channels=8, out_channels=8, kernel_size=(3, 3), groups = 8, padding=0, bias=False),\n",
        "            nn.Conv2d(in_channels=8, out_channels=12, kernel_size=(1, 1), groups = 1, padding=0, bias=False),\n",
        "            nn.ReLU(),\n",
        "            nn.BatchNorm2d(12),            \n",
        "        ) # input_size = 26 , output_size = 24, 16 to 32 channels\n",
        "\n",
        "        self.dropout1  = nn.Sequential(nn.Dropout(0.05))\n",
        "    #####################################################################################################\n",
        "\n",
        "        # TRANSITION BLOCK 1\n",
        "        self.convblock3 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=12, out_channels=8, kernel_size=(1, 1), padding=0, bias=False),\n",
        "        ) # input size  = 24, output_size = 24\n",
        "        self.pool1 = nn.MaxPool2d(2, 2) # input sizez = 24, output_size = 12\n",
        "    #####################################################################################################\n",
        "\n",
        "        # CONVOLUTION BLOCK 2\n",
        "        self.convblock4 = nn.Sequential(\n",
        "            # DEPTH WISE SEPARABLE CONVOLUTIONS\n",
        "            nn.Conv2d(in_channels=8, out_channels=8, kernel_size=(3, 3), groups = 8, padding=0, bias=False),\n",
        "            nn.Conv2d(in_channels=8, out_channels=12, kernel_size=(1, 1), padding=0, bias=False),\n",
        "            nn.ReLU(),            \n",
        "            nn.BatchNorm2d(12),\n",
        "            # nn.Dropout(dropout_value)\n",
        "        ) # input size = 12, output_size = 10\n",
        "        self.convblock5 = nn.Sequential(\n",
        "            # DEPTH WISE SEPARABLE CONVOLUTIONS\n",
        "            nn.Conv2d(in_channels=12, out_channels=12, kernel_size=(3, 3),groups = 12, padding=0, bias=False),\n",
        "            nn.Conv2d(in_channels=12, out_channels=16, kernel_size=(1, 1), padding=0, bias=False),\n",
        "            nn.ReLU(),            \n",
        "            nn.BatchNorm2d(16),\n",
        "            # nn.Dropout(dropout_value)\n",
        "        ) # input size = 10, output_size = 8\n",
        "        self.dropout2  = nn.Sequential(nn.Dropout(0.17))\n",
        "    #####################################################################################################\n",
        "\n",
        "        # adding a 1x1 kernel  block here to reduce paramters\n",
        "        self.convblock5a = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=16, out_channels=12, kernel_size=(1, 1), padding=0, bias=False),\n",
        "        ) # input size  = 8, output_size = 8\n",
        "        self.pool2 = nn.MaxPool2d(2, 2) # input sizez = 8, output_size = 4\n",
        "    #####################################################################################################\n",
        "\n",
        "        self.convblock6 = nn.Sequential(\n",
        "            # DEPTH WISE SEPARABLE CONVOLUTIONS\n",
        "            nn.Conv2d(in_channels=12, out_channels=12,groups = 12, kernel_size=(3, 3), padding=1, bias=False),\n",
        "            nn.Conv2d(in_channels=12, out_channels=15, kernel_size=(1, 1), padding=0, bias=False),\n",
        "            nn.ReLU(),            \n",
        "            nn.BatchNorm2d(15),\n",
        "            # nn.Dropout(dropout_value)\n",
        "        ) # input size  = 4,  output_size = 4\n",
        "        self.dropout3  = nn.Sequential(nn.Dropout(0.17))\n",
        "    ####################################### TRIAL BLOCK ##############################################################  \n",
        "        # self.convblock6 = nn.Sequential(\n",
        "        #     # DEPTH WISE SEPARABLE CONVOLUTIONS\n",
        "        #     nn.Conv2d(in_channels=12, out_channels=12,groups = 12, kernel_size=(3, 3), padding=1, bias=False),\n",
        "        #     nn.Conv2d(in_channels=12, out_channels=15, kernel_size=(1, 1), padding=0, bias=False),\n",
        "        #     nn.ReLU(),            \n",
        "        #     nn.BatchNorm2d(15),\n",
        "        #     # nn.Dropout(dropout_value)\n",
        "        # ) # input size  = 4,  output_size = 4\n",
        "        # self.dropout3  = nn.Sequential(nn.Dropout(0.17))\n",
        "\n",
        "        #####################################################################################################        \n",
        "        # OUTPUT BLOCK\n",
        "        self.gap = nn.Sequential(\n",
        "            nn.AvgPool2d(kernel_size=4)\n",
        "        ) # output_size = 1 channels  = 16\n",
        "        # self.flat = nn.Flatten()\n",
        "        self.convblock8 = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=15, out_channels=10, kernel_size=(1, 1), padding=0, bias=False),\n",
        "        ) \n",
        "        # self.linear = nn.Linear(15,10)\n",
        "\n",
        "    def forward(self, x):\n",
        "        # first conv block - student\n",
        "        # x = x.to(\"cuda\")\n",
        "        x = self.convblock1(x)\n",
        "        x = self.convblock2(x)\n",
        "        x  = self.dropout1(x)\n",
        "\n",
        "        # transition block\n",
        "        x = self.convblock3(x)\n",
        "        x = self.pool1(x)\n",
        "\n",
        "        # second conv block\n",
        "        x = self.convblock4(x)\n",
        "        x = self.convblock5(x)\n",
        "        x  = self.dropout2(x)\n",
        "\n",
        "        # secocnd transition block\n",
        "        x = self.convblock5a(x)\n",
        "        x = self.pool2(x)        \n",
        "\n",
        "        # third conv block\n",
        "        x = self.convblock6(x)\n",
        "        x = self.dropout3(x)\n",
        "        # x = self.convblock7(x)\n",
        "\n",
        "        # global average pooling layer\n",
        "        x = self.gap(x) \n",
        "        # x = self.flat(x)\n",
        "        # classification        \n",
        "        x = self.convblock8(x)\n",
        "        # x = self.linear(x)\n",
        "        x = x.view(-1, 10)\n",
        "        return F.log_softmax(x, dim=-1)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "GtrGVQUG2EvL"
      },
      "source": [
        "#### SMALLEST MODEL 2"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "uopsWOZRJGBD"
      },
      "source": [
        "class SimpleModel(composer.models.MosaicClassifier):\n",
        "    def __init__(self, num_classes: int):\n",
        "        module = nn.Sequential(\n",
        "    ################################## 1 ST CONVOLUTIONAL BLOCK #####################################  \n",
        "            nn.Conv2d(in_channels=1, out_channels=8, kernel_size=(3, 3), padding=0, bias=False), \n",
        "            nn.ReLU(),\n",
        "            nn.BatchNorm2d(8),                                                                                #26x26 /8 \n",
        "\n",
        "            nn.Conv2d(in_channels=8, out_channels=8, kernel_size=(3, 3), groups = 8, padding=0, bias=False),  #24x24/8\n",
        "            nn.Conv2d(in_channels=8, out_channels=12, kernel_size=(1, 1), groups = 1, padding=0, bias=False), #24x24/12\n",
        "            nn.ReLU(),\n",
        "            nn.BatchNorm2d(12), \n",
        "\n",
        "            nn.Dropout(0.05),\n",
        "    ################################### TRANSITION BLOCK #######################################\n",
        "            nn.Conv2d(in_channels=12, out_channels=8, kernel_size=(1, 1), padding=0, bias=False),             #24x24/8\n",
        "            nn.MaxPool2d(2, 2),                                                                               #12x12/8\n",
        "    ################################## 2 ND CONVOLUTIONAL BLOCK #####################################            \n",
        "\n",
        "            nn.Conv2d(in_channels=8, out_channels=8, kernel_size=(3, 3), groups = 8, padding=0, bias=False),  #10x10/8\n",
        "            nn.Conv2d(in_channels=8, out_channels=12, kernel_size=(1, 1), padding=0, bias=False),             #10x10/12\n",
        "            nn.ReLU(),            \n",
        "            nn.BatchNorm2d(12),\n",
        "\n",
        "            # DEPTH WISE SEPARABLE CONVOLUTIONS\n",
        "            nn.Conv2d(in_channels=12, out_channels=12, kernel_size=(3, 3),groups = 12, padding=0, bias=False), #8x8/12\n",
        "            nn.Conv2d(in_channels=12, out_channels=16, kernel_size=(1, 1), padding=0, bias=False),             #8x8/16\n",
        "            nn.ReLU(),            \n",
        "            nn.BatchNorm2d(16),\n",
        "            \n",
        "            nn.Dropout(0.17),\n",
        "    ################################### TRANSITION BLOCK #######################################\n",
        "            nn.Conv2d(in_channels=16, out_channels=12, kernel_size=(1, 1), padding=0, bias=False),            #8x8/12\n",
        "            nn.MaxPool2d(2, 2),                                                                               #4x4/12\n",
        "    ################################## 3 RD CONVOLUTIONAL BLOCK ##################################### \n",
        "            nn.Conv2d(in_channels=12, out_channels=12,groups = 12, kernel_size=(3, 3), padding=1, bias=False),#4x4/12\n",
        "            nn.Conv2d(in_channels=12, out_channels=15, kernel_size=(1, 1), padding=0, bias=False),            #4x4/15\n",
        "            nn.ReLU(),            \n",
        "            nn.BatchNorm2d(15),\n",
        "            \n",
        "            nn.Dropout(0.17),\n",
        "    ################################### OUTPUT BLOCK ######################################        \n",
        "            nn.AvgPool2d(kernel_size=4),                                                                      #1x1/15\n",
        "            # nn.Linear(1,10),\n",
        "            nn.Flatten(),\n",
        "            # nn.Conv2d(in_channels=15, out_channels=num_classes, kernel_size=(1, 1), padding=0, bias=False),\n",
        "            nn.Linear(15,10)\n",
        "        )\n",
        "        self.num_classes = num_classes\n",
        "        super().__init__(module=module)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "xBqPrS73IzAm"
      },
      "source": [
        "#### MODEL LOADING"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "XJZ2xk2vJTCa"
      },
      "source": [
        "# m = SimpleModel(num_classes= 10).to('cuda')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "6qVwtDegH-KK"
      },
      "source": [
        "m = SNet().to('cuda')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "IdcNWTTPQbcw",
        "outputId": "d82d017d-a737-4620-907a-a22ee264030b"
      },
      "source": [
        "summary(m, input_size=(1,28,28))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "----------------------------------------------------------------\n",
            "        Layer (type)               Output Shape         Param #\n",
            "================================================================\n",
            "            Conv2d-1            [-1, 8, 26, 26]              72\n",
            "              ReLU-2            [-1, 8, 26, 26]               0\n",
            "       BatchNorm2d-3            [-1, 8, 26, 26]              16\n",
            "            Conv2d-4            [-1, 8, 24, 24]              72\n",
            "            Conv2d-5           [-1, 12, 24, 24]              96\n",
            "              ReLU-6           [-1, 12, 24, 24]               0\n",
            "       BatchNorm2d-7           [-1, 12, 24, 24]              24\n",
            "           Dropout-8           [-1, 12, 24, 24]               0\n",
            "            Conv2d-9            [-1, 8, 24, 24]              96\n",
            "        MaxPool2d-10            [-1, 8, 12, 12]               0\n",
            "           Conv2d-11            [-1, 8, 10, 10]              72\n",
            "           Conv2d-12           [-1, 12, 10, 10]              96\n",
            "             ReLU-13           [-1, 12, 10, 10]               0\n",
            "      BatchNorm2d-14           [-1, 12, 10, 10]              24\n",
            "           Conv2d-15             [-1, 12, 8, 8]             108\n",
            "           Conv2d-16             [-1, 16, 8, 8]             192\n",
            "             ReLU-17             [-1, 16, 8, 8]               0\n",
            "      BatchNorm2d-18             [-1, 16, 8, 8]              32\n",
            "          Dropout-19             [-1, 16, 8, 8]               0\n",
            "           Conv2d-20             [-1, 12, 8, 8]             192\n",
            "        MaxPool2d-21             [-1, 12, 4, 4]               0\n",
            "           Conv2d-22             [-1, 12, 4, 4]             108\n",
            "           Conv2d-23             [-1, 15, 4, 4]             180\n",
            "             ReLU-24             [-1, 15, 4, 4]               0\n",
            "      BatchNorm2d-25             [-1, 15, 4, 4]              30\n",
            "          Dropout-26             [-1, 15, 4, 4]               0\n",
            "        AvgPool2d-27             [-1, 15, 1, 1]               0\n",
            "           Conv2d-28             [-1, 10, 1, 1]             150\n",
            "================================================================\n",
            "Total params: 1,560\n",
            "Trainable params: 1,560\n",
            "Non-trainable params: 0\n",
            "----------------------------------------------------------------\n",
            "Input size (MB): 0.00\n",
            "Forward/backward pass size (MB): 0.50\n",
            "Params size (MB): 0.01\n",
            "Estimated Total Size (MB): 0.51\n",
            "----------------------------------------------------------------\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "-2C0zMVIXNAY"
      },
      "source": [
        "#### MODEL AND DATASET LOADING\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "73pMWEMLImCI"
      },
      "source": [
        "use_cuda = True\n",
        "train_dataloader_spec = DataloaderSpec(\n",
        "    dataset=datasets.MNIST('/datasets/', train=True, transform=transforms.ToTensor(), download=True),\n",
        "    drop_last=False,\n",
        "    shuffle=True,\n",
        ")\n",
        "\n",
        "eval_dataloader_spec = DataloaderSpec(\n",
        "    dataset=datasets.MNIST('/datasets/', train=False, transform=transforms.ToTensor()),\n",
        "    drop_last=False,\n",
        "    shuffle=False,\n",
        ")\n",
        "\n",
        "\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "QpuE95hTJvT3",
        "outputId": "9967d152-c1d5-4811-8e1e-7495b378244f"
      },
      "source": [
        "use_cuda = True\n",
        "transforms = transforms.Compose([\n",
        "                                transforms.ToTensor(),\n",
        "                                transforms.Normalize((0.1307,), (0.3081,))\n",
        "                                ])\n",
        "train = datasets.MNIST('./data', train=True, download=True, transform=transforms)\n",
        "test = datasets.MNIST('./data', train=False, download=True, transform=transforms)\n",
        "dataloader_args = dict(shuffle=True, batch_size=128, num_workers=4, pin_memory=True) if use_cuda else dict(shuffle=True, batch_size=64)\n",
        "train_loader = torch.utils.data.DataLoader(train, **dataloader_args)\n",
        "test_loader = torch.utils.data.DataLoader(test, **dataloader_args)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/torch/utils/data/dataloader.py:481: UserWarning: This DataLoader will create 4 worker processes in total. Our suggested max number of worker in current system is 2, which is smaller than what this DataLoader is going to create. Please be aware that excessive worker creation might get DataLoader running slow or even freeze, lower the worker number to avoid potential slowness/freeze if necessary.\n",
            "  cpuset_checked))\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "use_cuda = True\n",
        "transforms = transforms.Compose([\n",
        "                                transforms.ToTensor(),\n",
        "                                transforms.Normalize((0.1307,), (0.3081,))\n",
        "                                ])\n",
        "train = datasets.CIFAR10('./data10', train=True, download=True, transform=transforms)\n",
        "test = datasets.CIFAR10('./data10', train=False, download=True, transform=transforms)\n",
        "\n",
        "dataloader_args = dict(shuffle=True, batch_size=128, num_workers=4, pin_memory=True) if use_cuda else dict(shuffle=True, batch_size=64)\n",
        "train_loader = torch.utils.data.DataLoader(train, **dataloader_args)\n",
        "test_loader = torch.utils.data.DataLoader(test, **dataloader_args)\n"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "z1dDabTp8dps",
        "outputId": "084c433c-e445-414d-f212-b9eacbfb2da1"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Files already downloaded and verified\n",
            "Files already downloaded and verified\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/torch/utils/data/dataloader.py:481: UserWarning: This DataLoader will create 4 worker processes in total. Our suggested max number of worker in current system is 2, which is smaller than what this DataLoader is going to create. Please be aware that excessive worker creation might get DataLoader running slow or even freeze, lower the worker number to avoid potential slowness/freeze if necessary.\n",
            "  cpuset_checked))\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "#### SUBSET DATALOADING"
      ],
      "metadata": {
        "id": "v5ad97Ce40Wk"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "use_cuda = True\n",
        "transforms = transforms.Compose([\n",
        "                                transforms.ToTensor(),\n",
        "                                transforms.Normalize((0.1307,), (0.3081,))\n",
        "                                ])\n",
        "train = datasets.MNIST('./data2', train=True, download=True, transform=transforms)\n",
        "test = datasets.MNIST('./data2', train=False, download=True, transform=transforms)\n",
        "\n",
        "train_subset = list(range(0, 45000))\n",
        "train_1 = torch.utils.data.Subset(train, train_subset)\n",
        "\n",
        "dataloader_args = dict(shuffle=True, batch_size=128, num_workers=4, pin_memory=True) if use_cuda else dict(shuffle=True, batch_size=64)\n",
        "train_loader = torch.utils.data.DataLoader(train_1, **dataloader_args)\n",
        "test_loader = torch.utils.data.DataLoader(test, **dataloader_args)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "6SoEBVaX4tzi",
        "outputId": "ef937917-6611-4d0d-94f7-aed8f8e73525"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/torch/utils/data/dataloader.py:481: UserWarning: This DataLoader will create 4 worker processes in total. Our suggested max number of worker in current system is 2, which is smaller than what this DataLoader is going to create. Please be aware that excessive worker creation might get DataLoader running slow or even freeze, lower the worker number to avoid potential slowness/freeze if necessary.\n",
            "  cpuset_checked))\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "LTgpVHkFnvZk"
      },
      "source": [
        "#### WEIGHT TRANSFER EXPERIMENTS"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "ygWCgwC5Gy5D",
        "outputId": "68e4d476-23e2-4056-b523-1c121367de93"
      },
      "source": [
        "# from collections import OrderedDict\n",
        "\n",
        "# path_ls = './label_smoothing_checkpoints/ep60.pt'\n",
        "# state_items = torch.load(path_ls)['state']['model']\n",
        "\n",
        "# new_state_dict = OrderedDict()\n",
        "# for x,y in state_items.items():\n",
        "#     name = x[7:]\n",
        "#     new_state_dict[name] = y\n",
        "\n",
        "# m.load_state_dict(new_state_dict)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "module.module.0.weight convblock1.0.weight\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "OA5wA1PaPKGn",
        "outputId": "c05388d1-fe61-47d3-af78-0b33f50b0a49"
      },
      "source": [
        "from collections import OrderedDict\n",
        "\n",
        "path_ls = './label_smoothing_checkpoints/ep60.pt'\n",
        "state_items = torch.load(path_ls)['state']['model']\n",
        "\n",
        "new_state_dict = OrderedDict()\n",
        "for (x,a) in zip(state_items.items(),m.state_dict()) :\n",
        "    # name = x[7:]\n",
        "    # new_state_dict[name] = y\n",
        "    # comp_module_name, _ =\n",
        "    saved_weights = x[1]\n",
        "    saved_module_name = x[0]  \n",
        "    act_module_name = a\n",
        "\n",
        "    new_state_dict[act_module_name] = saved_weights\n",
        "\n",
        "    # print(x[0].split(),a)\n",
        "    # break\n",
        "\n",
        "m.load_state_dict(new_state_dict)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<All keys matched successfully>"
            ]
          },
          "metadata": {},
          "execution_count": 26
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "8ByA4cGgUIhK"
      },
      "source": [
        "#### TRAIN FUNCTION"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "f5f9C465sZ28"
      },
      "source": [
        "def train(model):\n",
        "    # print('working')\n",
        "    model.train()\n",
        "    pbar = tqdm(train_loader)\n",
        "    device = 'cuda'\n",
        "    correct = 0\n",
        "    processed = 0\n",
        "    train_loss = 0\n",
        "    for batch_idx, (data, target) in enumerate(pbar):\n",
        "      data, target = data.to(device), target.to(device)\n",
        "      optimizer.zero_grad()\n",
        "      y_pred = model(data)\n",
        "      loss = F.nll_loss(y_pred, target)\n",
        "      loss.backward()\n",
        "      optimizer.step()\n",
        "      pred = y_pred.argmax(dim=1, keepdim=True)  # get the index of the max log-probability\n",
        "      correct += pred.eq(target.view_as(pred)).sum().item()\n",
        "      processed += len(data)\n",
        "      pbar.set_description(desc= f'Loss={loss.item()} Batch_id={batch_idx} Accuracy={100*correct/processed:0.2f}')\n",
        "      train_loss += loss\n",
        "    train_acc.append(100*correct/processed)\n",
        "    train_losses.append(train_loss)\n",
        "\n",
        "    # return self.train_acc, self.train_loss\n",
        "\n",
        "def test(model):\n",
        "    model.eval()\n",
        "    test_loss = 0\n",
        "    correct = 0\n",
        "    device = 'cuda'\n",
        "    with torch.no_grad():\n",
        "        for data, target in test_loader:\n",
        "            data, target = data.to(device), target.to(device)\n",
        "            output = model(data)\n",
        "            test_loss += F.nll_loss(output, target, reduction='sum').item()  # sum up batch loss\n",
        "            pred = output.argmax(dim=1, keepdim=True)  # get the index of the max log-probability\n",
        "            correct += pred.eq(target.view_as(pred)).sum().item()\n",
        "\n",
        "    test_loss /= len(test_loader.dataset)\n",
        "    test_losses.append(test_loss)\n",
        "\n",
        "    print('\\nTest set: Average loss: {:.4f}, Accuracy: {}/{} ({:.2f}%)\\n'.format(\n",
        "        test_loss, correct, len(test_loader.dataset),\n",
        "        100. * correct / len(test_loader.dataset)))\n",
        "\n",
        "    test_acc.append(100. * correct / len(test_loader.dataset))\n",
        "\n",
        "def execute_model(model,save_best_model = False):\n",
        "    for epoch in range(epochs):\n",
        "        # best_accuracy = []\n",
        "        \n",
        "        print(\"EPOCH:\", epoch)\n",
        "        train(model)\n",
        "        scheduler.step()\n",
        "        test(model)\n",
        "        if save_best_model == True:\n",
        "            best_accuracy = max(test_acc)\n",
        "            current_accuracy = test_acc[-1]\n",
        "            print(best_accuracy,current_accuracy)\n",
        "            if current_accuracy == best_accuracy or current_accuracy > best_accuracy:\n",
        "                torch.save(model.state_dict(), path)\n",
        "                print('Saving model here')\n",
        "\n",
        "def plot_curves(a,log = False):    \n",
        "    assert type(a) == str\n",
        "    plt.figure(figsize = (15,5))\n",
        "    plt.subplot(121)\n",
        "    plt.plot(train_acc)\n",
        "    plt.plot(test_acc)\n",
        "    if log == True:\n",
        "      plt.yscale('log')\n",
        "    plt.title(a+' Accuracy Curves')\n",
        "    plt.legend(['Train','Validation'])\n",
        "\n",
        "    plt.subplot(122)\n",
        "    plt.plot(train_losses)\n",
        "    plt.plot(test_losses)\n",
        "    if log == True:\n",
        "      plt.yscale('log')\n",
        "    plt.title(a+' Loss Curves')\n",
        "    plt.legend(['Train','Validation'])\n",
        "    plt.show()\n",
        "\n",
        "train_losses = []\n",
        "test_losses = []\n",
        "train_acc = []\n",
        "test_acc = []"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "#### TRAIN FUNCTION SUBSET"
      ],
      "metadata": {
        "id": "YtWMZyVG5BHq"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "def train(model):\n",
        "    # print('working')\n",
        "    model.train()\n",
        "    pbar = tqdm(train_loader)\n",
        "    correct = 0\n",
        "    processed = 0\n",
        "    train_loss = 0\n",
        "    for batch_idx, (data, target) in enumerate(pbar):\n",
        "      data, target = data.to(device), target.to(device)\n",
        "      optimizer.zero_grad()\n",
        "      y_pred = model(data)\n",
        "      loss = F.nll_loss(y_pred, target)\n",
        "      loss.backward()\n",
        "      optimizer.step()\n",
        "      pred = y_pred.argmax(dim=1, keepdim=True)  # get the index of the max log-probability\n",
        "      correct += pred.eq(target.view_as(pred)).sum().item()\n",
        "      processed += len(data)\n",
        "      pbar.set_description(desc= f'Loss={loss.item()} Batch_id={batch_idx} Accuracy={100*correct/processed:0.2f}')\n",
        "      train_loss += loss\n",
        "    optimizer.swap_swa_sgd() # extra code to add the stochaism\n",
        "    train_acc.append(100*correct/processed)\n",
        "    train_losses.append(train_loss)\n",
        "\n",
        "    # return self.train_acc, self.train_loss\n",
        "\n",
        "def test(model):\n",
        "    model.eval()\n",
        "    test_loss = 0\n",
        "    correct = 0\n",
        "    with torch.no_grad():\n",
        "        for data, target in test_loader:\n",
        "            data, target = data.to(device), target.to(device)\n",
        "            output = model(data)\n",
        "            test_loss += F.nll_loss(output, target, reduction='sum').item()  # sum up batch loss\n",
        "            pred = output.argmax(dim=1, keepdim=True)  # get the index of the max log-probability\n",
        "            correct += pred.eq(target.view_as(pred)).sum().item()\n",
        "\n",
        "    test_loss /= len(test_loader.dataset)\n",
        "    test_losses.append(test_loss)\n",
        "\n",
        "    print('\\nTest set: Average loss: {:.4f}, Accuracy: {}/{} ({:.2f}%)\\n'.format(\n",
        "        test_loss, correct, len(test_loader.dataset),\n",
        "        100. * correct / len(test_loader.dataset)))\n",
        "\n",
        "    test_acc.append(100. * correct / len(test_loader.dataset))\n",
        "\n",
        "def execute_model(model,save_best_model = False):\n",
        "    for epoch in range(epochs):\n",
        "        print(\"EPOCH:\", epoch)\n",
        "        train(model)\n",
        "        scheduler.step()\n",
        "        test(model)\n",
        "        if save_best_model == True:\n",
        "            best_accuracy = max(test_acc)\n",
        "            current_accuracy = test_acc[-1]\n",
        "            print(best_accuracy,current_accuracy)\n",
        "            if current_accuracy == best_accuracy or current_accuracy > best_accuracy:\n",
        "                torch.save(model.state_dict(), path)\n",
        "                print('Saving model here')\n",
        "\n",
        "def plot_curves(a,log = False):    \n",
        "    assert type(a) == str\n",
        "    plt.figure(figsize = (15,5))\n",
        "    plt.subplot(121)\n",
        "    plt.plot(train_acc)\n",
        "    plt.plot(test_acc)\n",
        "    if log == True:\n",
        "      plt.yscale('log')\n",
        "    plt.title(a+' Accuracy Curves')\n",
        "    plt.legend(['Train','Validation'])\n",
        "\n",
        "    plt.subplot(122)\n",
        "    plt.plot(train_losses)\n",
        "    plt.plot(test_losses)\n",
        "    if log == True:\n",
        "      plt.yscale('log')\n",
        "    plt.title(a+' Loss Curves')\n",
        "    plt.legend(['Train','Validation'])\n",
        "    plt.show()\n",
        "\n",
        "train_losses = []\n",
        "test_losses = []\n",
        "train_acc = []\n",
        "test_acc = []"
      ],
      "metadata": {
        "id": "Y9kaMea15C8T"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "epochs = 100\n",
        "device = 'cuda'\n",
        "opt = optim.Adam(m.parameters(), lr = 0.01)\n",
        "optimizer = SWA(opt, swa_start=50, swa_freq=3, swa_lr=0.05)\n",
        "scheduler = StepLR(optimizer, step_size=5, gamma=0.1)\n",
        "path = './quantization/pytorch_model/pyswa_mnist.pth'\n",
        "execute_model(m, save_best_model=True)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "NdpTH1bO5MUK",
        "outputId": "77649961-17ad-4317-ae82-777a8386ba6c"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "EPOCH: 0\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "\r  0%|          | 0/352 [00:00<?, ?it/s]/usr/local/lib/python3.7/dist-packages/torch/utils/data/dataloader.py:481: UserWarning: This DataLoader will create 4 worker processes in total. Our suggested max number of worker in current system is 2, which is smaller than what this DataLoader is going to create. Please be aware that excessive worker creation might get DataLoader running slow or even freeze, lower the worker number to avoid potential slowness/freeze if necessary.\n",
            "  cpuset_checked))\n",
            "Loss=0.13514575362205505 Batch_id=351 Accuracy=84.42: 100%|██████████| 352/352 [00:13<00:00, 25.30it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 6.1780, Accuracy: 1636/10000 (16.36%)\n",
            "\n",
            "16.36 16.36\n",
            "Saving model here\n",
            "EPOCH: 1\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.1623702347278595 Batch_id=351 Accuracy=93.98: 100%|██████████| 352/352 [00:13<00:00, 25.58it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 3.8750, Accuracy: 3714/10000 (37.14%)\n",
            "\n",
            "37.14 37.14\n",
            "Saving model here\n",
            "EPOCH: 2\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.08116363734006882 Batch_id=351 Accuracy=94.89: 100%|██████████| 352/352 [00:13<00:00, 25.41it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 7.3204, Accuracy: 1283/10000 (12.83%)\n",
            "\n",
            "37.14 12.83\n",
            "EPOCH: 3\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.08575272560119629 Batch_id=351 Accuracy=95.29: 100%|██████████| 352/352 [00:13<00:00, 25.45it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 4.7359, Accuracy: 2821/10000 (28.21%)\n",
            "\n",
            "37.14 28.21\n",
            "EPOCH: 4\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.10778459161520004 Batch_id=351 Accuracy=95.66: 100%|██████████| 352/352 [00:13<00:00, 25.39it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 1.0399, Accuracy: 6479/10000 (64.79%)\n",
            "\n",
            "64.79 64.79\n",
            "Saving model here\n",
            "EPOCH: 5\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.27189144492149353 Batch_id=351 Accuracy=95.96: 100%|██████████| 352/352 [00:15<00:00, 22.73it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 1.7713, Accuracy: 5011/10000 (50.11%)\n",
            "\n",
            "64.79 50.11\n",
            "EPOCH: 6\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.08003214001655579 Batch_id=351 Accuracy=95.98: 100%|██████████| 352/352 [00:14<00:00, 25.14it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.7718, Accuracy: 7325/10000 (73.25%)\n",
            "\n",
            "73.25 73.25\n",
            "Saving model here\n",
            "EPOCH: 7\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.20577695965766907 Batch_id=351 Accuracy=96.20: 100%|██████████| 352/352 [00:14<00:00, 25.03it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 6.6860, Accuracy: 1778/10000 (17.78%)\n",
            "\n",
            "73.25 17.78\n",
            "EPOCH: 8\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.06045835465192795 Batch_id=351 Accuracy=96.24: 100%|██████████| 352/352 [00:13<00:00, 25.20it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 2.5755, Accuracy: 2837/10000 (28.37%)\n",
            "\n",
            "73.25 28.37\n",
            "EPOCH: 9\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.3137270212173462 Batch_id=351 Accuracy=96.46: 100%|██████████| 352/352 [00:14<00:00, 25.12it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 5.2482, Accuracy: 2753/10000 (27.53%)\n",
            "\n",
            "73.25 27.53\n",
            "EPOCH: 10\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.11554417014122009 Batch_id=351 Accuracy=96.43: 100%|██████████| 352/352 [00:14<00:00, 24.93it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.9069, Accuracy: 7406/10000 (74.06%)\n",
            "\n",
            "74.06 74.06\n",
            "Saving model here\n",
            "EPOCH: 11\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.1886400580406189 Batch_id=351 Accuracy=96.67: 100%|██████████| 352/352 [00:14<00:00, 24.91it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.4323, Accuracy: 8588/10000 (85.88%)\n",
            "\n",
            "85.88 85.88\n",
            "Saving model here\n",
            "EPOCH: 12\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.0625610202550888 Batch_id=351 Accuracy=96.53: 100%|██████████| 352/352 [00:13<00:00, 25.15it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 1.6404, Accuracy: 5413/10000 (54.13%)\n",
            "\n",
            "85.88 54.13\n",
            "EPOCH: 13\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.1173611730337143 Batch_id=351 Accuracy=96.72: 100%|██████████| 352/352 [00:14<00:00, 24.75it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 1.3194, Accuracy: 6510/10000 (65.10%)\n",
            "\n",
            "85.88 65.1\n",
            "EPOCH: 14\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.0931302085518837 Batch_id=351 Accuracy=96.80: 100%|██████████| 352/352 [00:14<00:00, 24.98it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.5303, Accuracy: 8011/10000 (80.11%)\n",
            "\n",
            "85.88 80.11\n",
            "EPOCH: 15\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.10266578942537308 Batch_id=351 Accuracy=96.92: 100%|██████████| 352/352 [00:14<00:00, 24.96it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 1.7768, Accuracy: 6098/10000 (60.98%)\n",
            "\n",
            "85.88 60.98\n",
            "EPOCH: 16\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.015324986539781094 Batch_id=351 Accuracy=96.98: 100%|██████████| 352/352 [00:14<00:00, 24.68it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.3489, Accuracy: 8845/10000 (88.45%)\n",
            "\n",
            "88.45 88.45\n",
            "Saving model here\n",
            "EPOCH: 17\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.07101725041866302 Batch_id=351 Accuracy=96.92: 100%|██████████| 352/352 [00:14<00:00, 24.86it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 1.3337, Accuracy: 6701/10000 (67.01%)\n",
            "\n",
            "88.45 67.01\n",
            "EPOCH: 18\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.054743990302085876 Batch_id=351 Accuracy=97.02: 100%|██████████| 352/352 [00:14<00:00, 24.92it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 2.2164, Accuracy: 4393/10000 (43.93%)\n",
            "\n",
            "88.45 43.93\n",
            "EPOCH: 19\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.07801829278469086 Batch_id=351 Accuracy=97.00: 100%|██████████| 352/352 [00:14<00:00, 24.82it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.5336, Accuracy: 8450/10000 (84.50%)\n",
            "\n",
            "88.45 84.5\n",
            "EPOCH: 20\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.12755759060382843 Batch_id=351 Accuracy=97.04: 100%|██████████| 352/352 [00:14<00:00, 24.58it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.3549, Accuracy: 8860/10000 (88.60%)\n",
            "\n",
            "88.6 88.6\n",
            "Saving model here\n",
            "EPOCH: 21\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.046080734580755234 Batch_id=351 Accuracy=97.17: 100%|██████████| 352/352 [00:14<00:00, 24.79it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.5144, Accuracy: 8562/10000 (85.62%)\n",
            "\n",
            "88.6 85.62\n",
            "EPOCH: 22\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.054949887096881866 Batch_id=351 Accuracy=97.26: 100%|██████████| 352/352 [00:14<00:00, 24.86it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.8898, Accuracy: 7129/10000 (71.29%)\n",
            "\n",
            "88.6 71.29\n",
            "EPOCH: 23\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.09408777952194214 Batch_id=351 Accuracy=97.21: 100%|██████████| 352/352 [00:14<00:00, 24.66it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.1579, Accuracy: 9546/10000 (95.46%)\n",
            "\n",
            "95.46 95.46\n",
            "Saving model here\n",
            "EPOCH: 24\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.03556250035762787 Batch_id=351 Accuracy=97.35: 100%|██████████| 352/352 [00:14<00:00, 24.74it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.8115, Accuracy: 7384/10000 (73.84%)\n",
            "\n",
            "95.46 73.84\n",
            "EPOCH: 25\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.20202070474624634 Batch_id=351 Accuracy=97.17: 100%|██████████| 352/352 [00:14<00:00, 25.06it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.1832, Accuracy: 9459/10000 (94.59%)\n",
            "\n",
            "95.46 94.59\n",
            "EPOCH: 26\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.08788469433784485 Batch_id=351 Accuracy=97.29: 100%|██████████| 352/352 [00:14<00:00, 24.61it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 1.3559, Accuracy: 5866/10000 (58.66%)\n",
            "\n",
            "95.46 58.66\n",
            "EPOCH: 27\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.3287287652492523 Batch_id=351 Accuracy=97.27: 100%|██████████| 352/352 [00:14<00:00, 24.85it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.5596, Accuracy: 8626/10000 (86.26%)\n",
            "\n",
            "95.46 86.26\n",
            "EPOCH: 28\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.027930239215493202 Batch_id=351 Accuracy=97.40: 100%|██████████| 352/352 [00:14<00:00, 24.51it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.2336, Accuracy: 9278/10000 (92.78%)\n",
            "\n",
            "95.46 92.78\n",
            "EPOCH: 29\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.15756607055664062 Batch_id=351 Accuracy=97.48: 100%|██████████| 352/352 [00:14<00:00, 24.93it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.6903, Accuracy: 8331/10000 (83.31%)\n",
            "\n",
            "95.46 83.31\n",
            "EPOCH: 30\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.04519461840391159 Batch_id=351 Accuracy=97.23: 100%|██████████| 352/352 [00:14<00:00, 24.84it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.4721, Accuracy: 8561/10000 (85.61%)\n",
            "\n",
            "95.46 85.61\n",
            "EPOCH: 31\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.10053817182779312 Batch_id=351 Accuracy=97.32: 100%|██████████| 352/352 [00:14<00:00, 24.74it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.6684, Accuracy: 8068/10000 (80.68%)\n",
            "\n",
            "95.46 80.68\n",
            "EPOCH: 32\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.06853792071342468 Batch_id=351 Accuracy=97.44: 100%|██████████| 352/352 [00:14<00:00, 24.73it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.2641, Accuracy: 9158/10000 (91.58%)\n",
            "\n",
            "95.46 91.58\n",
            "EPOCH: 33\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.10778115689754486 Batch_id=351 Accuracy=97.58: 100%|██████████| 352/352 [00:14<00:00, 24.59it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.9801, Accuracy: 7542/10000 (75.42%)\n",
            "\n",
            "95.46 75.42\n",
            "EPOCH: 34\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.07457248121500015 Batch_id=351 Accuracy=97.61: 100%|██████████| 352/352 [00:14<00:00, 24.58it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.1427, Accuracy: 9539/10000 (95.39%)\n",
            "\n",
            "95.46 95.39\n",
            "EPOCH: 35\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.03665432706475258 Batch_id=351 Accuracy=97.37: 100%|██████████| 352/352 [00:14<00:00, 24.22it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.7187, Accuracy: 8028/10000 (80.28%)\n",
            "\n",
            "95.46 80.28\n",
            "EPOCH: 36\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.07666714489459991 Batch_id=351 Accuracy=97.56: 100%|██████████| 352/352 [00:14<00:00, 24.35it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.2778, Accuracy: 9155/10000 (91.55%)\n",
            "\n",
            "95.46 91.55\n",
            "EPOCH: 37\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.010019046254456043 Batch_id=351 Accuracy=97.49: 100%|██████████| 352/352 [00:14<00:00, 24.38it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.8692, Accuracy: 7407/10000 (74.07%)\n",
            "\n",
            "95.46 74.07\n",
            "EPOCH: 38\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.07366441935300827 Batch_id=351 Accuracy=97.54: 100%|██████████| 352/352 [00:14<00:00, 24.53it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0890, Accuracy: 9725/10000 (97.25%)\n",
            "\n",
            "97.25 97.25\n",
            "Saving model here\n",
            "EPOCH: 39\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.14884859323501587 Batch_id=351 Accuracy=97.56: 100%|██████████| 352/352 [00:14<00:00, 24.34it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 2.1767, Accuracy: 5470/10000 (54.70%)\n",
            "\n",
            "97.25 54.7\n",
            "EPOCH: 40\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.1300661414861679 Batch_id=351 Accuracy=97.57: 100%|██████████| 352/352 [00:14<00:00, 24.26it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.5911, Accuracy: 8067/10000 (80.67%)\n",
            "\n",
            "97.25 80.67\n",
            "EPOCH: 41\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.04224634915590286 Batch_id=351 Accuracy=97.67: 100%|██████████| 352/352 [00:14<00:00, 24.38it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.2087, Accuracy: 9315/10000 (93.15%)\n",
            "\n",
            "97.25 93.15\n",
            "EPOCH: 42\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.02031305991113186 Batch_id=351 Accuracy=97.56: 100%|██████████| 352/352 [00:14<00:00, 24.11it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.6458, Accuracy: 8132/10000 (81.32%)\n",
            "\n",
            "97.25 81.32\n",
            "EPOCH: 43\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.05565920099616051 Batch_id=351 Accuracy=97.67: 100%|██████████| 352/352 [00:14<00:00, 24.45it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 3.5121, Accuracy: 4362/10000 (43.62%)\n",
            "\n",
            "97.25 43.62\n",
            "EPOCH: 44\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.04264393448829651 Batch_id=351 Accuracy=97.68: 100%|██████████| 352/352 [00:14<00:00, 24.15it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 1.4022, Accuracy: 7107/10000 (71.07%)\n",
            "\n",
            "97.25 71.07\n",
            "EPOCH: 45\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.10058204084634781 Batch_id=351 Accuracy=97.71: 100%|██████████| 352/352 [00:14<00:00, 24.08it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.3516, Accuracy: 8979/10000 (89.79%)\n",
            "\n",
            "97.25 89.79\n",
            "EPOCH: 46\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.09006381779909134 Batch_id=351 Accuracy=97.66: 100%|██████████| 352/352 [00:14<00:00, 23.97it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.5359, Accuracy: 8286/10000 (82.86%)\n",
            "\n",
            "97.25 82.86\n",
            "EPOCH: 47\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.07265788316726685 Batch_id=351 Accuracy=97.65: 100%|██████████| 352/352 [00:14<00:00, 24.06it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.7580, Accuracy: 8063/10000 (80.63%)\n",
            "\n",
            "97.25 80.63\n",
            "EPOCH: 48\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.0830678716301918 Batch_id=351 Accuracy=97.65: 100%|██████████| 352/352 [00:14<00:00, 23.96it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.2869, Accuracy: 9002/10000 (90.02%)\n",
            "\n",
            "97.25 90.02\n",
            "EPOCH: 49\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.16221779584884644 Batch_id=351 Accuracy=97.75: 100%|██████████| 352/352 [00:14<00:00, 24.34it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.2009, Accuracy: 9424/10000 (94.24%)\n",
            "\n",
            "97.25 94.24\n",
            "EPOCH: 50\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.024730056524276733 Batch_id=351 Accuracy=97.64: 100%|██████████| 352/352 [00:14<00:00, 24.08it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.2029, Accuracy: 9373/10000 (93.73%)\n",
            "\n",
            "97.25 93.73\n",
            "EPOCH: 51\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.09183564782142639 Batch_id=351 Accuracy=97.78: 100%|██████████| 352/352 [00:14<00:00, 24.43it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.2893, Accuracy: 9176/10000 (91.76%)\n",
            "\n",
            "97.25 91.76\n",
            "EPOCH: 52\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.0725444033741951 Batch_id=351 Accuracy=97.59: 100%|██████████| 352/352 [00:14<00:00, 24.46it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.4588, Accuracy: 8545/10000 (85.45%)\n",
            "\n",
            "97.25 85.45\n",
            "EPOCH: 53\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.04898902773857117 Batch_id=351 Accuracy=97.68: 100%|██████████| 352/352 [00:14<00:00, 24.26it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.1279, Accuracy: 9639/10000 (96.39%)\n",
            "\n",
            "97.25 96.39\n",
            "EPOCH: 54\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.07743033766746521 Batch_id=351 Accuracy=97.87: 100%|██████████| 352/352 [00:14<00:00, 24.21it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.1375, Accuracy: 9576/10000 (95.76%)\n",
            "\n",
            "97.25 95.76\n",
            "EPOCH: 55\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.05263640359044075 Batch_id=351 Accuracy=97.81: 100%|██████████| 352/352 [00:14<00:00, 24.22it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.5395, Accuracy: 8573/10000 (85.73%)\n",
            "\n",
            "97.25 85.73\n",
            "EPOCH: 56\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.07455052435398102 Batch_id=351 Accuracy=97.83: 100%|██████████| 352/352 [00:14<00:00, 24.17it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.3651, Accuracy: 8915/10000 (89.15%)\n",
            "\n",
            "97.25 89.15\n",
            "EPOCH: 57\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.08463925868272781 Batch_id=351 Accuracy=97.82: 100%|██████████| 352/352 [00:14<00:00, 23.81it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.2581, Accuracy: 9166/10000 (91.66%)\n",
            "\n",
            "97.25 91.66\n",
            "EPOCH: 58\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.07204229384660721 Batch_id=351 Accuracy=97.76: 100%|██████████| 352/352 [00:14<00:00, 24.10it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.1572, Accuracy: 9513/10000 (95.13%)\n",
            "\n",
            "97.25 95.13\n",
            "EPOCH: 59\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.054816097021102905 Batch_id=351 Accuracy=97.88: 100%|██████████| 352/352 [00:14<00:00, 24.13it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.8478, Accuracy: 7756/10000 (77.56%)\n",
            "\n",
            "97.25 77.56\n",
            "EPOCH: 60\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.07496581226587296 Batch_id=351 Accuracy=97.80: 100%|██████████| 352/352 [00:14<00:00, 24.20it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.2863, Accuracy: 9061/10000 (90.61%)\n",
            "\n",
            "97.25 90.61\n",
            "EPOCH: 61\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.047827694565057755 Batch_id=351 Accuracy=97.86: 100%|██████████| 352/352 [00:14<00:00, 23.96it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.8793, Accuracy: 7848/10000 (78.48%)\n",
            "\n",
            "97.25 78.48\n",
            "EPOCH: 62\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.04814360663294792 Batch_id=351 Accuracy=97.72: 100%|██████████| 352/352 [00:15<00:00, 23.45it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.1693, Accuracy: 9471/10000 (94.71%)\n",
            "\n",
            "97.25 94.71\n",
            "EPOCH: 63\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.08441680669784546 Batch_id=351 Accuracy=97.92: 100%|██████████| 352/352 [00:15<00:00, 23.29it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.8365, Accuracy: 7710/10000 (77.10%)\n",
            "\n",
            "97.25 77.1\n",
            "EPOCH: 64\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.08790406584739685 Batch_id=351 Accuracy=97.90: 100%|██████████| 352/352 [00:15<00:00, 23.29it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.2356, Accuracy: 9233/10000 (92.33%)\n",
            "\n",
            "97.25 92.33\n",
            "EPOCH: 65\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.050615713000297546 Batch_id=351 Accuracy=97.85: 100%|██████████| 352/352 [00:14<00:00, 23.97it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.1268, Accuracy: 9631/10000 (96.31%)\n",
            "\n",
            "97.25 96.31\n",
            "EPOCH: 66\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.032613594084978104 Batch_id=351 Accuracy=97.90: 100%|██████████| 352/352 [00:14<00:00, 24.06it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.6588, Accuracy: 7974/10000 (79.74%)\n",
            "\n",
            "97.25 79.74\n",
            "EPOCH: 67\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.10210715234279633 Batch_id=351 Accuracy=97.99: 100%|██████████| 352/352 [00:14<00:00, 24.02it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.7646, Accuracy: 8106/10000 (81.06%)\n",
            "\n",
            "97.25 81.06\n",
            "EPOCH: 68\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.033275529742240906 Batch_id=351 Accuracy=97.97: 100%|██████████| 352/352 [00:14<00:00, 23.84it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.3187, Accuracy: 9019/10000 (90.19%)\n",
            "\n",
            "97.25 90.19\n",
            "EPOCH: 69\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.07921874523162842 Batch_id=351 Accuracy=98.00: 100%|██████████| 352/352 [00:14<00:00, 23.92it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.9416, Accuracy: 7715/10000 (77.15%)\n",
            "\n",
            "97.25 77.15\n",
            "EPOCH: 70\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.04017798602581024 Batch_id=351 Accuracy=97.90: 100%|██████████| 352/352 [00:14<00:00, 23.83it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.1466, Accuracy: 9532/10000 (95.32%)\n",
            "\n",
            "97.25 95.32\n",
            "EPOCH: 71\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.021118193864822388 Batch_id=351 Accuracy=97.90: 100%|██████████| 352/352 [00:14<00:00, 23.95it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.5749, Accuracy: 8480/10000 (84.80%)\n",
            "\n",
            "97.25 84.8\n",
            "EPOCH: 72\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.005150511395186186 Batch_id=351 Accuracy=97.99: 100%|██████████| 352/352 [00:14<00:00, 23.93it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.3939, Accuracy: 8738/10000 (87.38%)\n",
            "\n",
            "97.25 87.38\n",
            "EPOCH: 73\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.10197512805461884 Batch_id=351 Accuracy=97.90: 100%|██████████| 352/352 [00:14<00:00, 23.74it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.2993, Accuracy: 9140/10000 (91.40%)\n",
            "\n",
            "97.25 91.4\n",
            "EPOCH: 74\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.02314073033630848 Batch_id=351 Accuracy=97.90: 100%|██████████| 352/352 [00:14<00:00, 23.90it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.2323, Accuracy: 9268/10000 (92.68%)\n",
            "\n",
            "97.25 92.68\n",
            "EPOCH: 75\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.38897475600242615 Batch_id=351 Accuracy=98.01: 100%|██████████| 352/352 [00:14<00:00, 23.67it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.2050, Accuracy: 9399/10000 (93.99%)\n",
            "\n",
            "97.25 93.99\n",
            "EPOCH: 76\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.044514402747154236 Batch_id=351 Accuracy=97.91: 100%|██████████| 352/352 [00:14<00:00, 23.68it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.3383, Accuracy: 8930/10000 (89.30%)\n",
            "\n",
            "97.25 89.3\n",
            "EPOCH: 77\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.11744502931833267 Batch_id=351 Accuracy=97.99: 100%|██████████| 352/352 [00:14<00:00, 23.70it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.3565, Accuracy: 8942/10000 (89.42%)\n",
            "\n",
            "97.25 89.42\n",
            "EPOCH: 78\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.09893939644098282 Batch_id=351 Accuracy=97.97: 100%|██████████| 352/352 [00:14<00:00, 23.88it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.2551, Accuracy: 9199/10000 (91.99%)\n",
            "\n",
            "97.25 91.99\n",
            "EPOCH: 79\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.09653875231742859 Batch_id=351 Accuracy=98.05: 100%|██████████| 352/352 [00:14<00:00, 23.68it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.3078, Accuracy: 9168/10000 (91.68%)\n",
            "\n",
            "97.25 91.68\n",
            "EPOCH: 80\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.03034301847219467 Batch_id=351 Accuracy=97.99: 100%|██████████| 352/352 [00:14<00:00, 23.76it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.7335, Accuracy: 7773/10000 (77.73%)\n",
            "\n",
            "97.25 77.73\n",
            "EPOCH: 81\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.03589803725481033 Batch_id=351 Accuracy=98.11: 100%|██████████| 352/352 [00:14<00:00, 23.77it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.5391, Accuracy: 8657/10000 (86.57%)\n",
            "\n",
            "97.25 86.57\n",
            "EPOCH: 82\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.020419826731085777 Batch_id=351 Accuracy=98.11: 100%|██████████| 352/352 [00:14<00:00, 23.93it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.3214, Accuracy: 8909/10000 (89.09%)\n",
            "\n",
            "97.25 89.09\n",
            "EPOCH: 83\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.012614483013749123 Batch_id=351 Accuracy=97.89: 100%|██████████| 352/352 [00:14<00:00, 23.84it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.2041, Accuracy: 9468/10000 (94.68%)\n",
            "\n",
            "97.25 94.68\n",
            "EPOCH: 84\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.05410466715693474 Batch_id=351 Accuracy=98.06: 100%|██████████| 352/352 [00:14<00:00, 23.85it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.4707, Accuracy: 8516/10000 (85.16%)\n",
            "\n",
            "97.25 85.16\n",
            "EPOCH: 85\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.0870797336101532 Batch_id=351 Accuracy=97.98: 100%|██████████| 352/352 [00:14<00:00, 23.85it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.3493, Accuracy: 9164/10000 (91.64%)\n",
            "\n",
            "97.25 91.64\n",
            "EPOCH: 86\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.1137290894985199 Batch_id=351 Accuracy=98.01: 100%|██████████| 352/352 [00:14<00:00, 23.68it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 1.4303, Accuracy: 6340/10000 (63.40%)\n",
            "\n",
            "97.25 63.4\n",
            "EPOCH: 87\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.047384683042764664 Batch_id=351 Accuracy=97.98: 100%|██████████| 352/352 [00:14<00:00, 24.04it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.5051, Accuracy: 8644/10000 (86.44%)\n",
            "\n",
            "97.25 86.44\n",
            "EPOCH: 88\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.01025333534926176 Batch_id=351 Accuracy=98.00: 100%|██████████| 352/352 [00:14<00:00, 23.79it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.1942, Accuracy: 9399/10000 (93.99%)\n",
            "\n",
            "97.25 93.99\n",
            "EPOCH: 89\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.11579852551221848 Batch_id=351 Accuracy=98.07: 100%|██████████| 352/352 [00:14<00:00, 24.06it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.4697, Accuracy: 8877/10000 (88.77%)\n",
            "\n",
            "97.25 88.77\n",
            "EPOCH: 90\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.060796961188316345 Batch_id=351 Accuracy=98.04: 100%|██████████| 352/352 [00:14<00:00, 23.74it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.2352, Accuracy: 9273/10000 (92.73%)\n",
            "\n",
            "97.25 92.73\n",
            "EPOCH: 91\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.1006179228425026 Batch_id=351 Accuracy=98.17: 100%|██████████| 352/352 [00:15<00:00, 23.41it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.1760, Accuracy: 9486/10000 (94.86%)\n",
            "\n",
            "97.25 94.86\n",
            "EPOCH: 92\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.011275280267000198 Batch_id=351 Accuracy=98.10: 100%|██████████| 352/352 [00:14<00:00, 23.71it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.2749, Accuracy: 9128/10000 (91.28%)\n",
            "\n",
            "97.25 91.28\n",
            "EPOCH: 93\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.05025400593876839 Batch_id=351 Accuracy=98.03: 100%|██████████| 352/352 [00:15<00:00, 23.45it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.3066, Accuracy: 9221/10000 (92.21%)\n",
            "\n",
            "97.25 92.21\n",
            "EPOCH: 94\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.0693429484963417 Batch_id=351 Accuracy=98.00: 100%|██████████| 352/352 [00:14<00:00, 23.53it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.4112, Accuracy: 8776/10000 (87.76%)\n",
            "\n",
            "97.25 87.76\n",
            "EPOCH: 95\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.018897399306297302 Batch_id=351 Accuracy=98.16: 100%|██████████| 352/352 [00:14<00:00, 23.62it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.2164, Accuracy: 9390/10000 (93.90%)\n",
            "\n",
            "97.25 93.9\n",
            "EPOCH: 96\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.006646326277405024 Batch_id=351 Accuracy=98.09: 100%|██████████| 352/352 [00:14<00:00, 23.93it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.1291, Accuracy: 9569/10000 (95.69%)\n",
            "\n",
            "97.25 95.69\n",
            "EPOCH: 97\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.09079921990633011 Batch_id=351 Accuracy=98.11: 100%|██████████| 352/352 [00:14<00:00, 23.65it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.4824, Accuracy: 8669/10000 (86.69%)\n",
            "\n",
            "97.25 86.69\n",
            "EPOCH: 98\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.05752250924706459 Batch_id=351 Accuracy=97.95: 100%|██████████| 352/352 [00:14<00:00, 23.68it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.2240, Accuracy: 9288/10000 (92.88%)\n",
            "\n",
            "97.25 92.88\n",
            "EPOCH: 99\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.035918623208999634 Batch_id=351 Accuracy=98.19: 100%|██████████| 352/352 [00:14<00:00, 23.55it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 3.0992, Accuracy: 5779/10000 (57.79%)\n",
            "\n",
            "97.25 57.79\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "JbXR1-NKsS1e"
      },
      "source": [
        "#### 7K WITH DW TRAINING WITH EVALUATION"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "rSbNZ5MISuHS",
        "outputId": "da3a616f-0eb2-4c0a-c907-a7426be53dee"
      },
      "source": [
        "time.time()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "1638769815.752612"
            ]
          },
          "metadata": {},
          "execution_count": 16
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Iv0ilf-JsWoZ"
      },
      "source": [
        "# RUNNING THE INFERENCEC TIME ON A CPU\n",
        "test_losses = []\n",
        "test_acc =[]\n",
        "def test1(model):\n",
        "    model.eval()\n",
        "    test_loss = 0\n",
        "    correct = 0\n",
        "    # device = 'cuda'\n",
        "    time_list = []\n",
        "    with torch.no_grad():\n",
        "        for data, target in test_loader:\n",
        "            # data, target = data.to(device), target.to(device)\n",
        "            start = time.time()\n",
        "            output = model(data)\n",
        "            end = time.time()\n",
        "            interval = end-start\n",
        "            time_list.append(interval)\n",
        "            test_loss += F.nll_loss(output, target, reduction='sum').item()  # sum up batch loss\n",
        "            pred = output.argmax(dim=1, keepdim=True)  # get the index of the max log-probability\n",
        "            correct += pred.eq(target.view_as(pred)).sum().item()\n",
        "\n",
        "    test_loss /= len(test_loader.dataset)\n",
        "    test_losses.append(test_loss)\n",
        "    total_time = sum(time_list)\n",
        "    print('\\nTest set: Average loss: {:.4f}, Accuracy: {}/{} ({:.2f}%)\\n'.format(\n",
        "        test_loss, correct, len(test_loader.dataset),\n",
        "        100. * correct / len(test_loader.dataset)))\n",
        "    print(f'time taken is {total_time}')\n",
        "\n",
        "    test_acc.append(100. * correct / len(test_loader.dataset))"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 394
        },
        "id": "l47gKmN3MmPT",
        "outputId": "3efbb0ac-5a34-465c-c4b6-bdd0309d3cae"
      },
      "source": [
        "test_model = Net().to('cuda')\n",
        "summary(test_model, input_size = (1,28,28))\n",
        "\n",
        "epochs = 50\n",
        "optimizer = optim.Adam(test_model.parameters(), lr = 0.01)\n",
        "scheduler = StepLR(optimizer, step_size=5, gamma=0.1)\n",
        "path = './quantization/pytorch_model/mosaic_7k_dw.pth'\n",
        "\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "error",
          "ename": "RuntimeError",
          "evalue": "ignored",
          "traceback": [
            "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
            "\u001b[0;31mRuntimeError\u001b[0m                              Traceback (most recent call last)",
            "\u001b[0;32m<ipython-input-9-98d612a146ff>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mtest_model\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mNet\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mto\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m'cuda'\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      2\u001b[0m \u001b[0msummary\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mtest_model\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0minput_size\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m(\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;36m28\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;36m28\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      3\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      4\u001b[0m \u001b[0mepochs\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;36m50\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      5\u001b[0m \u001b[0moptimizer\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0moptim\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mAdam\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mtest_model\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mparameters\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mlr\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;36m0.01\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py\u001b[0m in \u001b[0;36mto\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m    897\u001b[0m             \u001b[0;32mreturn\u001b[0m \u001b[0mt\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mto\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mdevice\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mdtype\u001b[0m \u001b[0;32mif\u001b[0m \u001b[0mt\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mis_floating_point\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;32mor\u001b[0m \u001b[0mt\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mis_complex\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;32melse\u001b[0m \u001b[0;32mNone\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mnon_blocking\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    898\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 899\u001b[0;31m         \u001b[0;32mreturn\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_apply\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mconvert\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    900\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    901\u001b[0m     def register_backward_hook(\n",
            "\u001b[0;32m/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py\u001b[0m in \u001b[0;36m_apply\u001b[0;34m(self, fn)\u001b[0m\n\u001b[1;32m    568\u001b[0m     \u001b[0;32mdef\u001b[0m \u001b[0m_apply\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfn\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    569\u001b[0m         \u001b[0;32mfor\u001b[0m \u001b[0mmodule\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mchildren\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 570\u001b[0;31m             \u001b[0mmodule\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_apply\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mfn\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    571\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    572\u001b[0m         \u001b[0;32mdef\u001b[0m \u001b[0mcompute_should_use_set_data\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mtensor\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mtensor_applied\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py\u001b[0m in \u001b[0;36m_apply\u001b[0;34m(self, fn)\u001b[0m\n\u001b[1;32m    568\u001b[0m     \u001b[0;32mdef\u001b[0m \u001b[0m_apply\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfn\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    569\u001b[0m         \u001b[0;32mfor\u001b[0m \u001b[0mmodule\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mchildren\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 570\u001b[0;31m             \u001b[0mmodule\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_apply\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mfn\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    571\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    572\u001b[0m         \u001b[0;32mdef\u001b[0m \u001b[0mcompute_should_use_set_data\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mtensor\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mtensor_applied\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py\u001b[0m in \u001b[0;36m_apply\u001b[0;34m(self, fn)\u001b[0m\n\u001b[1;32m    591\u001b[0m             \u001b[0;31m# `with torch.no_grad():`\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    592\u001b[0m             \u001b[0;32mwith\u001b[0m \u001b[0mtorch\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mno_grad\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 593\u001b[0;31m                 \u001b[0mparam_applied\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mfn\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mparam\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    594\u001b[0m             \u001b[0mshould_use_set_data\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mcompute_should_use_set_data\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mparam\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mparam_applied\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    595\u001b[0m             \u001b[0;32mif\u001b[0m \u001b[0mshould_use_set_data\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py\u001b[0m in \u001b[0;36mconvert\u001b[0;34m(t)\u001b[0m\n\u001b[1;32m    895\u001b[0m                 return t.to(device, dtype if t.is_floating_point() or t.is_complex() else None,\n\u001b[1;32m    896\u001b[0m                             non_blocking, memory_format=convert_to_format)\n\u001b[0;32m--> 897\u001b[0;31m             \u001b[0;32mreturn\u001b[0m \u001b[0mt\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mto\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mdevice\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mdtype\u001b[0m \u001b[0;32mif\u001b[0m \u001b[0mt\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mis_floating_point\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;32mor\u001b[0m \u001b[0mt\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mis_complex\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;32melse\u001b[0m \u001b[0;32mNone\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mnon_blocking\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    898\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    899\u001b[0m         \u001b[0;32mreturn\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_apply\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mconvert\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.7/dist-packages/torch/cuda/__init__.py\u001b[0m in \u001b[0;36m_lazy_init\u001b[0;34m()\u001b[0m\n\u001b[1;32m    212\u001b[0m         \u001b[0;31m# This function throws if there's a driver initialization error, no GPUs\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    213\u001b[0m         \u001b[0;31m# are found or any other error occurs\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 214\u001b[0;31m         \u001b[0mtorch\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_C\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_cuda_init\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    215\u001b[0m         \u001b[0;31m# Some of the queued calls may reentrantly call _lazy_init();\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    216\u001b[0m         \u001b[0;31m# we need to just return without initializing in that case.\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;31mRuntimeError\u001b[0m: No CUDA GPUs are available"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "-t47-lB3oNd1",
        "outputId": "9794a87b-f597-4971-ba32-2c859f495e24"
      },
      "source": [
        "execute_model(test_model, save_best_model=True)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "EPOCH: 0\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "\r  0%|          | 0/469 [00:00<?, ?it/s]/usr/local/lib/python3.7/dist-packages/torch/utils/data/dataloader.py:481: UserWarning: This DataLoader will create 4 worker processes in total. Our suggested max number of worker in current system is 2, which is smaller than what this DataLoader is going to create. Please be aware that excessive worker creation might get DataLoader running slow or even freeze, lower the worker number to avoid potential slowness/freeze if necessary.\n",
            "  cpuset_checked))\n",
            "Loss=0.1082153394818306 Batch_id=468 Accuracy=93.60: 100%|██████████| 469/469 [00:12<00:00, 37.05it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0614, Accuracy: 9818/10000 (98.18%)\n",
            "\n",
            "98.18 98.18\n",
            "Saving model here\n",
            "EPOCH: 1\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.02054537832736969 Batch_id=468 Accuracy=97.80: 100%|██████████| 469/469 [00:12<00:00, 36.97it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0555, Accuracy: 9814/10000 (98.14%)\n",
            "\n",
            "98.18 98.14\n",
            "EPOCH: 2\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.0875418409705162 Batch_id=468 Accuracy=98.18: 100%|██████████| 469/469 [00:12<00:00, 37.21it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0368, Accuracy: 9879/10000 (98.79%)\n",
            "\n",
            "98.79 98.79\n",
            "Saving model here\n",
            "EPOCH: 3\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.02730736881494522 Batch_id=468 Accuracy=98.37: 100%|██████████| 469/469 [00:12<00:00, 36.97it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0384, Accuracy: 9884/10000 (98.84%)\n",
            "\n",
            "98.84 98.84\n",
            "Saving model here\n",
            "EPOCH: 4\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.025228897109627724 Batch_id=468 Accuracy=98.53: 100%|██████████| 469/469 [00:12<00:00, 36.85it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0416, Accuracy: 9865/10000 (98.65%)\n",
            "\n",
            "98.84 98.65\n",
            "EPOCH: 5\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.06000388041138649 Batch_id=468 Accuracy=99.06: 100%|██████████| 469/469 [00:12<00:00, 37.38it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0217, Accuracy: 9934/10000 (99.34%)\n",
            "\n",
            "99.34 99.34\n",
            "Saving model here\n",
            "EPOCH: 6\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.001279215794056654 Batch_id=468 Accuracy=99.13: 100%|██████████| 469/469 [00:12<00:00, 37.53it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0211, Accuracy: 9935/10000 (99.35%)\n",
            "\n",
            "99.35 99.35\n",
            "Saving model here\n",
            "EPOCH: 7\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.014887847937643528 Batch_id=468 Accuracy=99.16: 100%|██████████| 469/469 [00:12<00:00, 37.44it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0216, Accuracy: 9932/10000 (99.32%)\n",
            "\n",
            "99.35 99.32\n",
            "EPOCH: 8\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.13382670283317566 Batch_id=468 Accuracy=99.26: 100%|██████████| 469/469 [00:12<00:00, 36.73it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0199, Accuracy: 9934/10000 (99.34%)\n",
            "\n",
            "99.35 99.34\n",
            "EPOCH: 9\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.08737765997648239 Batch_id=468 Accuracy=99.27: 100%|██████████| 469/469 [00:12<00:00, 36.89it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0190, Accuracy: 9942/10000 (99.42%)\n",
            "\n",
            "99.42 99.42\n",
            "Saving model here\n",
            "EPOCH: 10\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.050947513431310654 Batch_id=468 Accuracy=99.34: 100%|██████████| 469/469 [00:12<00:00, 37.47it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0184, Accuracy: 9944/10000 (99.44%)\n",
            "\n",
            "99.44 99.44\n",
            "Saving model here\n",
            "EPOCH: 11\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.08939076215028763 Batch_id=468 Accuracy=99.31: 100%|██████████| 469/469 [00:12<00:00, 37.56it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0184, Accuracy: 9940/10000 (99.40%)\n",
            "\n",
            "99.44 99.4\n",
            "EPOCH: 12\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.060498252511024475 Batch_id=468 Accuracy=99.36: 100%|██████████| 469/469 [00:12<00:00, 36.91it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0186, Accuracy: 9943/10000 (99.43%)\n",
            "\n",
            "99.44 99.43\n",
            "EPOCH: 13\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.015667391940951347 Batch_id=468 Accuracy=99.27: 100%|██████████| 469/469 [00:12<00:00, 36.91it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0181, Accuracy: 9942/10000 (99.42%)\n",
            "\n",
            "99.44 99.42\n",
            "EPOCH: 14\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.07250838726758957 Batch_id=468 Accuracy=99.38: 100%|██████████| 469/469 [00:12<00:00, 36.85it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0184, Accuracy: 9944/10000 (99.44%)\n",
            "\n",
            "99.44 99.44\n",
            "Saving model here\n",
            "EPOCH: 15\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.018782032653689384 Batch_id=468 Accuracy=99.34: 100%|██████████| 469/469 [00:12<00:00, 36.83it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0183, Accuracy: 9943/10000 (99.43%)\n",
            "\n",
            "99.44 99.43\n",
            "EPOCH: 16\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.011412869207561016 Batch_id=468 Accuracy=99.33: 100%|██████████| 469/469 [00:12<00:00, 36.26it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0184, Accuracy: 9942/10000 (99.42%)\n",
            "\n",
            "99.44 99.42\n",
            "EPOCH: 17\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.004648513626307249 Batch_id=468 Accuracy=99.43: 100%|██████████| 469/469 [00:12<00:00, 36.15it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0189, Accuracy: 9943/10000 (99.43%)\n",
            "\n",
            "99.44 99.43\n",
            "EPOCH: 18\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.017759809270501137 Batch_id=468 Accuracy=99.36: 100%|██████████| 469/469 [00:13<00:00, 35.96it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0185, Accuracy: 9943/10000 (99.43%)\n",
            "\n",
            "99.44 99.43\n",
            "EPOCH: 19\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.04964422807097435 Batch_id=468 Accuracy=99.32: 100%|██████████| 469/469 [00:12<00:00, 36.18it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0184, Accuracy: 9941/10000 (99.41%)\n",
            "\n",
            "99.44 99.41\n",
            "EPOCH: 20\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.04003634676337242 Batch_id=468 Accuracy=99.39: 100%|██████████| 469/469 [00:12<00:00, 36.41it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0182, Accuracy: 9945/10000 (99.45%)\n",
            "\n",
            "99.45 99.45\n",
            "Saving model here\n",
            "EPOCH: 21\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.05418994650244713 Batch_id=468 Accuracy=99.32: 100%|██████████| 469/469 [00:12<00:00, 36.70it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0191, Accuracy: 9944/10000 (99.44%)\n",
            "\n",
            "99.45 99.44\n",
            "EPOCH: 22\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.03063797391951084 Batch_id=468 Accuracy=99.36: 100%|██████████| 469/469 [00:12<00:00, 36.40it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0185, Accuracy: 9944/10000 (99.44%)\n",
            "\n",
            "99.45 99.44\n",
            "EPOCH: 23\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.008022727444767952 Batch_id=468 Accuracy=99.42: 100%|██████████| 469/469 [00:12<00:00, 36.43it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0190, Accuracy: 9942/10000 (99.42%)\n",
            "\n",
            "99.45 99.42\n",
            "EPOCH: 24\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.010450811125338078 Batch_id=468 Accuracy=99.36: 100%|██████████| 469/469 [00:12<00:00, 36.52it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0182, Accuracy: 9944/10000 (99.44%)\n",
            "\n",
            "99.45 99.44\n",
            "EPOCH: 25\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.031612563878297806 Batch_id=468 Accuracy=99.38: 100%|██████████| 469/469 [00:12<00:00, 36.77it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0180, Accuracy: 9945/10000 (99.45%)\n",
            "\n",
            "99.45 99.45\n",
            "Saving model here\n",
            "EPOCH: 26\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.059387292712926865 Batch_id=468 Accuracy=99.39: 100%|██████████| 469/469 [00:13<00:00, 35.96it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0183, Accuracy: 9944/10000 (99.44%)\n",
            "\n",
            "99.45 99.44\n",
            "EPOCH: 27\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.017665257677435875 Batch_id=468 Accuracy=99.38: 100%|██████████| 469/469 [00:12<00:00, 36.36it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0182, Accuracy: 9943/10000 (99.43%)\n",
            "\n",
            "99.45 99.43\n",
            "EPOCH: 28\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.012305264361202717 Batch_id=468 Accuracy=99.34: 100%|██████████| 469/469 [00:12<00:00, 36.26it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0187, Accuracy: 9943/10000 (99.43%)\n",
            "\n",
            "99.45 99.43\n",
            "EPOCH: 29\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.0024973589461296797 Batch_id=468 Accuracy=99.43: 100%|██████████| 469/469 [00:12<00:00, 36.82it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0179, Accuracy: 9944/10000 (99.44%)\n",
            "\n",
            "99.45 99.44\n",
            "EPOCH: 30\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.027332833036780357 Batch_id=468 Accuracy=99.39: 100%|██████████| 469/469 [00:12<00:00, 37.02it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0181, Accuracy: 9944/10000 (99.44%)\n",
            "\n",
            "99.45 99.44\n",
            "EPOCH: 31\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.007543714251369238 Batch_id=468 Accuracy=99.39: 100%|██████████| 469/469 [00:12<00:00, 36.87it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0178, Accuracy: 9946/10000 (99.46%)\n",
            "\n",
            "99.46 99.46\n",
            "Saving model here\n",
            "EPOCH: 32\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.01399508398026228 Batch_id=468 Accuracy=99.40: 100%|██████████| 469/469 [00:12<00:00, 36.87it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0180, Accuracy: 9944/10000 (99.44%)\n",
            "\n",
            "99.46 99.44\n",
            "EPOCH: 33\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.00539654865860939 Batch_id=468 Accuracy=99.38: 100%|██████████| 469/469 [00:12<00:00, 37.08it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0181, Accuracy: 9944/10000 (99.44%)\n",
            "\n",
            "99.46 99.44\n",
            "EPOCH: 34\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.014303344301879406 Batch_id=468 Accuracy=99.38: 100%|██████████| 469/469 [00:12<00:00, 36.82it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0189, Accuracy: 9943/10000 (99.43%)\n",
            "\n",
            "99.46 99.43\n",
            "EPOCH: 35\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.004373711068183184 Batch_id=468 Accuracy=99.37: 100%|██████████| 469/469 [00:13<00:00, 35.93it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0184, Accuracy: 9944/10000 (99.44%)\n",
            "\n",
            "99.46 99.44\n",
            "EPOCH: 36\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.0016808713553473353 Batch_id=468 Accuracy=99.36: 100%|██████████| 469/469 [00:12<00:00, 36.13it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0187, Accuracy: 9941/10000 (99.41%)\n",
            "\n",
            "99.46 99.41\n",
            "EPOCH: 37\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.00405402947217226 Batch_id=468 Accuracy=99.38: 100%|██████████| 469/469 [00:13<00:00, 35.15it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0189, Accuracy: 9944/10000 (99.44%)\n",
            "\n",
            "99.46 99.44\n",
            "EPOCH: 38\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.005847516935318708 Batch_id=468 Accuracy=99.37: 100%|██████████| 469/469 [00:13<00:00, 35.08it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0184, Accuracy: 9942/10000 (99.42%)\n",
            "\n",
            "99.46 99.42\n",
            "EPOCH: 39\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.004465402569621801 Batch_id=468 Accuracy=99.40: 100%|██████████| 469/469 [00:13<00:00, 35.91it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0189, Accuracy: 9942/10000 (99.42%)\n",
            "\n",
            "99.46 99.42\n",
            "EPOCH: 40\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.032483313232660294 Batch_id=468 Accuracy=99.40: 100%|██████████| 469/469 [00:13<00:00, 35.75it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0184, Accuracy: 9942/10000 (99.42%)\n",
            "\n",
            "99.46 99.42\n",
            "EPOCH: 41\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.014879980124533176 Batch_id=468 Accuracy=99.35: 100%|██████████| 469/469 [00:13<00:00, 35.38it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0180, Accuracy: 9946/10000 (99.46%)\n",
            "\n",
            "99.46 99.46\n",
            "Saving model here\n",
            "EPOCH: 42\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.018985634669661522 Batch_id=468 Accuracy=99.33: 100%|██████████| 469/469 [00:13<00:00, 35.75it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0183, Accuracy: 9946/10000 (99.46%)\n",
            "\n",
            "99.46 99.46\n",
            "Saving model here\n",
            "EPOCH: 43\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.002994045615196228 Batch_id=468 Accuracy=99.35: 100%|██████████| 469/469 [00:13<00:00, 35.72it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0181, Accuracy: 9943/10000 (99.43%)\n",
            "\n",
            "99.46 99.43\n",
            "EPOCH: 44\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.00778867257758975 Batch_id=468 Accuracy=99.36: 100%|██████████| 469/469 [00:13<00:00, 35.45it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0191, Accuracy: 9941/10000 (99.41%)\n",
            "\n",
            "99.46 99.41\n",
            "EPOCH: 45\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.002465047175064683 Batch_id=468 Accuracy=99.33: 100%|██████████| 469/469 [00:13<00:00, 35.18it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0180, Accuracy: 9945/10000 (99.45%)\n",
            "\n",
            "99.46 99.45\n",
            "EPOCH: 46\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.005003333557397127 Batch_id=468 Accuracy=99.38: 100%|██████████| 469/469 [00:13<00:00, 35.84it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0183, Accuracy: 9944/10000 (99.44%)\n",
            "\n",
            "99.46 99.44\n",
            "EPOCH: 47\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.008983730338513851 Batch_id=468 Accuracy=99.39: 100%|██████████| 469/469 [00:13<00:00, 34.64it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0181, Accuracy: 9944/10000 (99.44%)\n",
            "\n",
            "99.46 99.44\n",
            "EPOCH: 48\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.033340808004140854 Batch_id=468 Accuracy=99.39: 100%|██████████| 469/469 [00:13<00:00, 35.58it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0189, Accuracy: 9941/10000 (99.41%)\n",
            "\n",
            "99.46 99.41\n",
            "EPOCH: 49\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.002529869554564357 Batch_id=468 Accuracy=99.38: 100%|██████████| 469/469 [00:13<00:00, 35.99it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0181, Accuracy: 9944/10000 (99.44%)\n",
            "\n",
            "99.46 99.44\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 336
        },
        "id": "bEoPWPFzr046",
        "outputId": "5be448f0-907d-49b7-f842-ef43483482df"
      },
      "source": [
        "plot_curves('7k params with dw',test_model)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1080x360 with 2 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "BoReBChtcZxC"
      },
      "source": [
        "#### 7K MODEL WITH 7.6 K PARAMS\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "f7vED-haWid5",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "6aed134f-3541-40a5-87cb-46855ec9385f"
      },
      "source": [
        "test_model = Net1().to('cuda')\n",
        "summary(test_model, input_size = (1,28,28))\n",
        "\n",
        "epochs = 50\n",
        "optimizer = optim.Adam(test_model.parameters(), lr = 0.01)\n",
        "scheduler = StepLR(optimizer, step_size=5, gamma=0.1)\n",
        "path = './quantization/pytorch_model/mosaic_7k_dw_new.pth'\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "----------------------------------------------------------------\n",
            "        Layer (type)               Output Shape         Param #\n",
            "================================================================\n",
            "            Conv2d-1           [-1, 10, 26, 26]              90\n",
            "              ReLU-2           [-1, 10, 26, 26]               0\n",
            "       BatchNorm2d-3           [-1, 10, 26, 26]              20\n",
            "            Conv2d-4           [-1, 10, 24, 24]              90\n",
            "            Conv2d-5           [-1, 12, 24, 24]             120\n",
            "              ReLU-6           [-1, 12, 24, 24]               0\n",
            "       BatchNorm2d-7           [-1, 12, 24, 24]              24\n",
            "           Dropout-8           [-1, 12, 24, 24]               0\n",
            "            Conv2d-9           [-1, 10, 24, 24]             120\n",
            "        MaxPool2d-10           [-1, 10, 12, 12]               0\n",
            "           Conv2d-11           [-1, 13, 10, 10]           1,170\n",
            "             ReLU-12           [-1, 13, 10, 10]               0\n",
            "      BatchNorm2d-13           [-1, 13, 10, 10]              26\n",
            "           Conv2d-14             [-1, 16, 8, 8]           1,872\n",
            "             ReLU-15             [-1, 16, 8, 8]               0\n",
            "      BatchNorm2d-16             [-1, 16, 8, 8]              32\n",
            "          Dropout-17             [-1, 16, 8, 8]               0\n",
            "           Conv2d-18             [-1, 12, 8, 8]             192\n",
            "        MaxPool2d-19             [-1, 12, 4, 4]               0\n",
            "           Conv2d-20             [-1, 15, 4, 4]           1,620\n",
            "             ReLU-21             [-1, 15, 4, 4]               0\n",
            "      BatchNorm2d-22             [-1, 15, 4, 4]              30\n",
            "          Dropout-23             [-1, 15, 4, 4]               0\n",
            "           Conv2d-24             [-1, 15, 4, 4]           2,025\n",
            "             ReLU-25             [-1, 15, 4, 4]               0\n",
            "      BatchNorm2d-26             [-1, 15, 4, 4]              30\n",
            "        AvgPool2d-27             [-1, 15, 1, 1]               0\n",
            "           Conv2d-28             [-1, 10, 1, 1]             150\n",
            "================================================================\n",
            "Total params: 7,611\n",
            "Trainable params: 7,611\n",
            "Non-trainable params: 0\n",
            "----------------------------------------------------------------\n",
            "Input size (MB): 0.00\n",
            "Forward/backward pass size (MB): 0.55\n",
            "Params size (MB): 0.03\n",
            "Estimated Total Size (MB): 0.58\n",
            "----------------------------------------------------------------\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "AtvxtTdGcPN-",
        "outputId": "7dbc9436-ecea-421e-d6e8-fe91fb7fa261"
      },
      "source": [
        "execute_model(test_model, save_best_model=True)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "EPOCH: 0\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "\r  0%|          | 0/469 [00:00<?, ?it/s]/usr/local/lib/python3.7/dist-packages/torch/utils/data/dataloader.py:481: UserWarning: This DataLoader will create 4 worker processes in total. Our suggested max number of worker in current system is 2, which is smaller than what this DataLoader is going to create. Please be aware that excessive worker creation might get DataLoader running slow or even freeze, lower the worker number to avoid potential slowness/freeze if necessary.\n",
            "  cpuset_checked))\n",
            "Loss=0.13090483844280243 Batch_id=468 Accuracy=93.51: 100%|██████████| 469/469 [00:11<00:00, 40.62it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.1281, Accuracy: 9607/10000 (96.07%)\n",
            "\n",
            "96.07 96.07\n",
            "Saving model here\n",
            "EPOCH: 1\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.06565754115581512 Batch_id=468 Accuracy=97.55: 100%|██████████| 469/469 [00:11<00:00, 40.87it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0567, Accuracy: 9813/10000 (98.13%)\n",
            "\n",
            "98.13 98.13\n",
            "Saving model here\n",
            "EPOCH: 2\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.031689006835222244 Batch_id=468 Accuracy=98.03: 100%|██████████| 469/469 [00:11<00:00, 40.28it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0488, Accuracy: 9849/10000 (98.49%)\n",
            "\n",
            "98.49 98.49\n",
            "Saving model here\n",
            "EPOCH: 3\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.175955668091774 Batch_id=468 Accuracy=98.27: 100%|██████████| 469/469 [00:11<00:00, 40.37it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0382, Accuracy: 9887/10000 (98.87%)\n",
            "\n",
            "98.87 98.87\n",
            "Saving model here\n",
            "EPOCH: 4\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.07044311612844467 Batch_id=468 Accuracy=98.41: 100%|██████████| 469/469 [00:11<00:00, 40.12it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0393, Accuracy: 9879/10000 (98.79%)\n",
            "\n",
            "98.87 98.79\n",
            "EPOCH: 5\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.008538560010492802 Batch_id=468 Accuracy=98.98: 100%|██████████| 469/469 [00:11<00:00, 40.44it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0258, Accuracy: 9921/10000 (99.21%)\n",
            "\n",
            "99.21 99.21\n",
            "Saving model here\n",
            "EPOCH: 6\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.041385192424058914 Batch_id=468 Accuracy=99.09: 100%|██████████| 469/469 [00:11<00:00, 40.64it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0253, Accuracy: 9911/10000 (99.11%)\n",
            "\n",
            "99.21 99.11\n",
            "EPOCH: 7\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.018815508112311363 Batch_id=468 Accuracy=99.08: 100%|██████████| 469/469 [00:11<00:00, 40.67it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0246, Accuracy: 9922/10000 (99.22%)\n",
            "\n",
            "99.22 99.22\n",
            "Saving model here\n",
            "EPOCH: 8\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.004526353906840086 Batch_id=468 Accuracy=99.12: 100%|██████████| 469/469 [00:11<00:00, 40.54it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0237, Accuracy: 9919/10000 (99.19%)\n",
            "\n",
            "99.22 99.19\n",
            "EPOCH: 9\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.003062059171497822 Batch_id=468 Accuracy=99.14: 100%|██████████| 469/469 [00:11<00:00, 40.23it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0247, Accuracy: 9925/10000 (99.25%)\n",
            "\n",
            "99.25 99.25\n",
            "Saving model here\n",
            "EPOCH: 10\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.008385790511965752 Batch_id=468 Accuracy=99.26: 100%|██████████| 469/469 [00:11<00:00, 40.52it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0236, Accuracy: 9924/10000 (99.24%)\n",
            "\n",
            "99.25 99.24\n",
            "EPOCH: 11\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.08226577192544937 Batch_id=468 Accuracy=99.28: 100%|██████████| 469/469 [00:11<00:00, 40.32it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0236, Accuracy: 9925/10000 (99.25%)\n",
            "\n",
            "99.25 99.25\n",
            "Saving model here\n",
            "EPOCH: 12\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.009610921144485474 Batch_id=468 Accuracy=99.24: 100%|██████████| 469/469 [00:11<00:00, 40.54it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0241, Accuracy: 9920/10000 (99.20%)\n",
            "\n",
            "99.25 99.2\n",
            "EPOCH: 13\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.002387305023148656 Batch_id=468 Accuracy=99.27: 100%|██████████| 469/469 [00:11<00:00, 40.78it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0232, Accuracy: 9928/10000 (99.28%)\n",
            "\n",
            "99.28 99.28\n",
            "Saving model here\n",
            "EPOCH: 14\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.006180583033710718 Batch_id=468 Accuracy=99.28: 100%|██████████| 469/469 [00:11<00:00, 40.39it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0230, Accuracy: 9928/10000 (99.28%)\n",
            "\n",
            "99.28 99.28\n",
            "Saving model here\n",
            "EPOCH: 15\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.007284469902515411 Batch_id=468 Accuracy=99.25: 100%|██████████| 469/469 [00:11<00:00, 40.09it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0236, Accuracy: 9927/10000 (99.27%)\n",
            "\n",
            "99.28 99.27\n",
            "EPOCH: 16\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.05470110476016998 Batch_id=468 Accuracy=99.31: 100%|██████████| 469/469 [00:11<00:00, 40.71it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0228, Accuracy: 9931/10000 (99.31%)\n",
            "\n",
            "99.31 99.31\n",
            "Saving model here\n",
            "EPOCH: 17\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.006730444263666868 Batch_id=468 Accuracy=99.29: 100%|██████████| 469/469 [00:11<00:00, 40.21it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0236, Accuracy: 9925/10000 (99.25%)\n",
            "\n",
            "99.31 99.25\n",
            "EPOCH: 18\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.008591183461248875 Batch_id=468 Accuracy=99.29: 100%|██████████| 469/469 [00:11<00:00, 40.79it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0245, Accuracy: 9926/10000 (99.26%)\n",
            "\n",
            "99.31 99.26\n",
            "EPOCH: 19\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.06954629719257355 Batch_id=468 Accuracy=99.28: 100%|██████████| 469/469 [00:11<00:00, 40.77it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0234, Accuracy: 9927/10000 (99.27%)\n",
            "\n",
            "99.31 99.27\n",
            "EPOCH: 20\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.013369708321988583 Batch_id=468 Accuracy=99.27: 100%|██████████| 469/469 [00:11<00:00, 40.96it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0232, Accuracy: 9926/10000 (99.26%)\n",
            "\n",
            "99.31 99.26\n",
            "EPOCH: 21\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.012492910958826542 Batch_id=468 Accuracy=99.28: 100%|██████████| 469/469 [00:11<00:00, 40.84it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0234, Accuracy: 9930/10000 (99.30%)\n",
            "\n",
            "99.31 99.3\n",
            "EPOCH: 22\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.023412197828292847 Batch_id=468 Accuracy=99.29: 100%|██████████| 469/469 [00:11<00:00, 40.67it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0230, Accuracy: 9927/10000 (99.27%)\n",
            "\n",
            "99.31 99.27\n",
            "EPOCH: 23\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.004855922888964415 Batch_id=468 Accuracy=99.31: 100%|██████████| 469/469 [00:12<00:00, 38.03it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0232, Accuracy: 9928/10000 (99.28%)\n",
            "\n",
            "99.31 99.28\n",
            "EPOCH: 24\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.005448854994028807 Batch_id=468 Accuracy=99.31: 100%|██████████| 469/469 [00:12<00:00, 38.91it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0228, Accuracy: 9927/10000 (99.27%)\n",
            "\n",
            "99.31 99.27\n",
            "EPOCH: 25\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.005849994253367186 Batch_id=468 Accuracy=99.31: 100%|██████████| 469/469 [00:12<00:00, 38.14it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0229, Accuracy: 9929/10000 (99.29%)\n",
            "\n",
            "99.31 99.29\n",
            "EPOCH: 26\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.006535604130476713 Batch_id=468 Accuracy=99.33: 100%|██████████| 469/469 [00:11<00:00, 39.73it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0235, Accuracy: 9925/10000 (99.25%)\n",
            "\n",
            "99.31 99.25\n",
            "EPOCH: 27\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.00594748929142952 Batch_id=468 Accuracy=99.28: 100%|██████████| 469/469 [00:11<00:00, 40.24it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0237, Accuracy: 9928/10000 (99.28%)\n",
            "\n",
            "99.31 99.28\n",
            "EPOCH: 28\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.0019165356643497944 Batch_id=468 Accuracy=99.26: 100%|██████████| 469/469 [00:11<00:00, 40.33it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0232, Accuracy: 9927/10000 (99.27%)\n",
            "\n",
            "99.31 99.27\n",
            "EPOCH: 29\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.006686723325401545 Batch_id=468 Accuracy=99.30: 100%|██████████| 469/469 [00:11<00:00, 39.75it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0239, Accuracy: 9925/10000 (99.25%)\n",
            "\n",
            "99.31 99.25\n",
            "EPOCH: 30\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.13173054158687592 Batch_id=468 Accuracy=99.27: 100%|██████████| 469/469 [00:11<00:00, 40.16it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0232, Accuracy: 9926/10000 (99.26%)\n",
            "\n",
            "99.31 99.26\n",
            "EPOCH: 31\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.018181422725319862 Batch_id=468 Accuracy=99.27: 100%|██████████| 469/469 [00:11<00:00, 39.79it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0238, Accuracy: 9924/10000 (99.24%)\n",
            "\n",
            "99.31 99.24\n",
            "EPOCH: 32\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.004015788435935974 Batch_id=468 Accuracy=99.27: 100%|██████████| 469/469 [00:11<00:00, 39.77it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0233, Accuracy: 9928/10000 (99.28%)\n",
            "\n",
            "99.31 99.28\n",
            "EPOCH: 33\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.08537200093269348 Batch_id=468 Accuracy=99.27: 100%|██████████| 469/469 [00:11<00:00, 39.13it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0233, Accuracy: 9929/10000 (99.29%)\n",
            "\n",
            "99.31 99.29\n",
            "EPOCH: 34\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.020938977599143982 Batch_id=468 Accuracy=99.28: 100%|██████████| 469/469 [00:12<00:00, 38.63it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0230, Accuracy: 9926/10000 (99.26%)\n",
            "\n",
            "99.31 99.26\n",
            "EPOCH: 35\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.02247251570224762 Batch_id=468 Accuracy=99.31: 100%|██████████| 469/469 [00:11<00:00, 39.71it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0240, Accuracy: 9926/10000 (99.26%)\n",
            "\n",
            "99.31 99.26\n",
            "EPOCH: 36\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.012650396674871445 Batch_id=468 Accuracy=99.29: 100%|██████████| 469/469 [00:11<00:00, 39.95it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0248, Accuracy: 9925/10000 (99.25%)\n",
            "\n",
            "99.31 99.25\n",
            "EPOCH: 37\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.03469930216670036 Batch_id=468 Accuracy=99.23: 100%|██████████| 469/469 [00:11<00:00, 40.06it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0242, Accuracy: 9925/10000 (99.25%)\n",
            "\n",
            "99.31 99.25\n",
            "EPOCH: 38\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.047761380672454834 Batch_id=468 Accuracy=99.29: 100%|██████████| 469/469 [00:11<00:00, 40.28it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0236, Accuracy: 9925/10000 (99.25%)\n",
            "\n",
            "99.31 99.25\n",
            "EPOCH: 39\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.012024745345115662 Batch_id=468 Accuracy=99.31: 100%|██████████| 469/469 [00:11<00:00, 39.54it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0232, Accuracy: 9925/10000 (99.25%)\n",
            "\n",
            "99.31 99.25\n",
            "EPOCH: 40\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.028406566008925438 Batch_id=468 Accuracy=99.31: 100%|██████████| 469/469 [00:11<00:00, 39.55it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0236, Accuracy: 9927/10000 (99.27%)\n",
            "\n",
            "99.31 99.27\n",
            "EPOCH: 41\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.008627836592495441 Batch_id=468 Accuracy=99.33: 100%|██████████| 469/469 [00:11<00:00, 39.66it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0229, Accuracy: 9930/10000 (99.30%)\n",
            "\n",
            "99.31 99.3\n",
            "EPOCH: 42\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.015173579566180706 Batch_id=468 Accuracy=99.27: 100%|██████████| 469/469 [00:11<00:00, 39.91it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0233, Accuracy: 9928/10000 (99.28%)\n",
            "\n",
            "99.31 99.28\n",
            "EPOCH: 43\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.015255617909133434 Batch_id=468 Accuracy=99.28: 100%|██████████| 469/469 [00:11<00:00, 40.27it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0232, Accuracy: 9927/10000 (99.27%)\n",
            "\n",
            "99.31 99.27\n",
            "EPOCH: 44\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.006850117817521095 Batch_id=468 Accuracy=99.28: 100%|██████████| 469/469 [00:11<00:00, 39.82it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0235, Accuracy: 9927/10000 (99.27%)\n",
            "\n",
            "99.31 99.27\n",
            "EPOCH: 45\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.007182084023952484 Batch_id=468 Accuracy=99.31: 100%|██████████| 469/469 [00:11<00:00, 39.45it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0228, Accuracy: 9933/10000 (99.33%)\n",
            "\n",
            "99.33 99.33\n",
            "Saving model here\n",
            "EPOCH: 46\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.0022405392955988646 Batch_id=468 Accuracy=99.28: 100%|██████████| 469/469 [00:11<00:00, 39.31it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0237, Accuracy: 9927/10000 (99.27%)\n",
            "\n",
            "99.33 99.27\n",
            "EPOCH: 47\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.002883447101339698 Batch_id=468 Accuracy=99.30: 100%|██████████| 469/469 [00:11<00:00, 39.10it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0228, Accuracy: 9930/10000 (99.30%)\n",
            "\n",
            "99.33 99.3\n",
            "EPOCH: 48\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.009814745746552944 Batch_id=468 Accuracy=99.26: 100%|██████████| 469/469 [00:11<00:00, 40.21it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0235, Accuracy: 9926/10000 (99.26%)\n",
            "\n",
            "99.33 99.26\n",
            "EPOCH: 49\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.045463334769010544 Batch_id=468 Accuracy=99.27: 100%|██████████| 469/469 [00:11<00:00, 39.49it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0235, Accuracy: 9926/10000 (99.26%)\n",
            "\n",
            "99.33 99.26\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 336
        },
        "id": "lwkPcfuPeOk_",
        "outputId": "c14cc524-b488-474c-96ec-332a218f9f05"
      },
      "source": [
        "plot_curves('7.6 k params with dw',test_model)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1080x360 with 2 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Cy63ZKBOthEF"
      },
      "source": [
        "#### 5.6K MODEL WITH DEPTHWISE CONV "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "2Hq_W3cktgca",
        "outputId": "98d4fd46-4b96-4242-ffbe-d4afd5a663fb"
      },
      "source": [
        "test_model = CNet1().to('cuda')\n",
        "summary(test_model, input_size = (1,28,28))\n",
        "\n",
        "epochs = 50\n",
        "optimizer = optim.Adam(test_model.parameters(), lr = 0.01)\n",
        "scheduler = StepLR(optimizer, step_size=5, gamma=0.1)\n",
        "path = './quantization/pytorch_model/mosaic_5K_dw_new.pth'"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "----------------------------------------------------------------\n",
            "        Layer (type)               Output Shape         Param #\n",
            "================================================================\n",
            "            Conv2d-1            [-1, 8, 26, 26]              72\n",
            "              ReLU-2            [-1, 8, 26, 26]               0\n",
            "       BatchNorm2d-3            [-1, 8, 26, 26]              16\n",
            "            Conv2d-4            [-1, 8, 24, 24]              72\n",
            "            Conv2d-5           [-1, 12, 24, 24]              96\n",
            "              ReLU-6           [-1, 12, 24, 24]               0\n",
            "       BatchNorm2d-7           [-1, 12, 24, 24]              24\n",
            "           Dropout-8           [-1, 12, 24, 24]               0\n",
            "            Conv2d-9            [-1, 8, 24, 24]              96\n",
            "        MaxPool2d-10            [-1, 8, 12, 12]               0\n",
            "           Conv2d-11           [-1, 13, 10, 10]             936\n",
            "             ReLU-12           [-1, 13, 10, 10]               0\n",
            "      BatchNorm2d-13           [-1, 13, 10, 10]              26\n",
            "           Conv2d-14             [-1, 18, 8, 8]           2,106\n",
            "             ReLU-15             [-1, 18, 8, 8]               0\n",
            "      BatchNorm2d-16             [-1, 18, 8, 8]              36\n",
            "          Dropout-17             [-1, 18, 8, 8]               0\n",
            "           Conv2d-18             [-1, 12, 8, 8]             216\n",
            "        MaxPool2d-19             [-1, 12, 4, 4]               0\n",
            "           Conv2d-20             [-1, 16, 4, 4]           1,728\n",
            "             ReLU-21             [-1, 16, 4, 4]               0\n",
            "      BatchNorm2d-22             [-1, 16, 4, 4]              32\n",
            "          Dropout-23             [-1, 16, 4, 4]               0\n",
            "        AvgPool2d-24             [-1, 16, 1, 1]               0\n",
            "           Conv2d-25             [-1, 10, 1, 1]             160\n",
            "================================================================\n",
            "Total params: 5,616\n",
            "Trainable params: 5,616\n",
            "Non-trainable params: 0\n",
            "----------------------------------------------------------------\n",
            "Input size (MB): 0.00\n",
            "Forward/backward pass size (MB): 0.49\n",
            "Params size (MB): 0.02\n",
            "Estimated Total Size (MB): 0.52\n",
            "----------------------------------------------------------------\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Zb-ymw5ztrGE",
        "outputId": "2d4245e9-bb08-4f89-e43d-fadb5fb80362"
      },
      "source": [
        "execute_model(test_model, save_best_model=True)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "EPOCH: 0\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "  0%|          | 0/469 [00:00<?, ?it/s]/usr/local/lib/python3.7/dist-packages/torch/utils/data/dataloader.py:481: UserWarning: This DataLoader will create 4 worker processes in total. Our suggested max number of worker in current system is 2, which is smaller than what this DataLoader is going to create. Please be aware that excessive worker creation might get DataLoader running slow or even freeze, lower the worker number to avoid potential slowness/freeze if necessary.\n",
            "  cpuset_checked))\n",
            "Loss=0.13055835664272308 Batch_id=468 Accuracy=92.44: 100%|██████████| 469/469 [00:11<00:00, 39.51it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.1017, Accuracy: 9651/10000 (96.51%)\n",
            "\n",
            "96.51 96.51\n",
            "Saving model here\n",
            "EPOCH: 1\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.04787528142333031 Batch_id=468 Accuracy=97.67: 100%|██████████| 469/469 [00:12<00:00, 38.89it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0436, Accuracy: 9868/10000 (98.68%)\n",
            "\n",
            "98.68 98.68\n",
            "Saving model here\n",
            "EPOCH: 2\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.019855692982673645 Batch_id=468 Accuracy=97.95: 100%|██████████| 469/469 [00:11<00:00, 39.74it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0483, Accuracy: 9838/10000 (98.38%)\n",
            "\n",
            "98.68 98.38\n",
            "EPOCH: 3\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.057521283626556396 Batch_id=468 Accuracy=98.25: 100%|██████████| 469/469 [00:11<00:00, 40.15it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0451, Accuracy: 9859/10000 (98.59%)\n",
            "\n",
            "98.68 98.59\n",
            "EPOCH: 4\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.11048400402069092 Batch_id=468 Accuracy=98.37: 100%|██████████| 469/469 [00:11<00:00, 39.38it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0370, Accuracy: 9882/10000 (98.82%)\n",
            "\n",
            "98.82 98.82\n",
            "Saving model here\n",
            "EPOCH: 5\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.0676373764872551 Batch_id=468 Accuracy=98.89: 100%|██████████| 469/469 [00:11<00:00, 39.89it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0262, Accuracy: 9914/10000 (99.14%)\n",
            "\n",
            "99.14 99.14\n",
            "Saving model here\n",
            "EPOCH: 6\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.04417817294597626 Batch_id=468 Accuracy=98.97: 100%|██████████| 469/469 [00:11<00:00, 39.40it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0234, Accuracy: 9927/10000 (99.27%)\n",
            "\n",
            "99.27 99.27\n",
            "Saving model here\n",
            "EPOCH: 7\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.05187194049358368 Batch_id=468 Accuracy=99.05: 100%|██████████| 469/469 [00:11<00:00, 39.51it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0260, Accuracy: 9924/10000 (99.24%)\n",
            "\n",
            "99.27 99.24\n",
            "EPOCH: 8\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.002177296206355095 Batch_id=468 Accuracy=99.09: 100%|██████████| 469/469 [00:11<00:00, 39.69it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0240, Accuracy: 9927/10000 (99.27%)\n",
            "\n",
            "99.27 99.27\n",
            "Saving model here\n",
            "EPOCH: 9\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.013960727490484715 Batch_id=468 Accuracy=99.08: 100%|██████████| 469/469 [00:11<00:00, 39.95it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0250, Accuracy: 9920/10000 (99.20%)\n",
            "\n",
            "99.27 99.2\n",
            "EPOCH: 10\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.012264303863048553 Batch_id=468 Accuracy=99.17: 100%|██████████| 469/469 [00:11<00:00, 39.12it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0237, Accuracy: 9927/10000 (99.27%)\n",
            "\n",
            "99.27 99.27\n",
            "Saving model here\n",
            "EPOCH: 11\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.02797161042690277 Batch_id=468 Accuracy=99.16: 100%|██████████| 469/469 [00:12<00:00, 38.49it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0231, Accuracy: 9929/10000 (99.29%)\n",
            "\n",
            "99.29 99.29\n",
            "Saving model here\n",
            "EPOCH: 12\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.04378479719161987 Batch_id=468 Accuracy=99.20: 100%|██████████| 469/469 [00:11<00:00, 39.10it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0235, Accuracy: 9925/10000 (99.25%)\n",
            "\n",
            "99.29 99.25\n",
            "EPOCH: 13\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.018713654950261116 Batch_id=468 Accuracy=99.19: 100%|██████████| 469/469 [00:11<00:00, 39.35it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0227, Accuracy: 9926/10000 (99.26%)\n",
            "\n",
            "99.29 99.26\n",
            "EPOCH: 14\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.026468550786376 Batch_id=468 Accuracy=99.21: 100%|██████████| 469/469 [00:11<00:00, 39.77it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0229, Accuracy: 9927/10000 (99.27%)\n",
            "\n",
            "99.29 99.27\n",
            "EPOCH: 15\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.009634794667363167 Batch_id=468 Accuracy=99.24: 100%|██████████| 469/469 [00:11<00:00, 39.68it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0227, Accuracy: 9926/10000 (99.26%)\n",
            "\n",
            "99.29 99.26\n",
            "EPOCH: 16\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.011882253922522068 Batch_id=468 Accuracy=99.15: 100%|██████████| 469/469 [00:11<00:00, 39.42it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0220, Accuracy: 9930/10000 (99.30%)\n",
            "\n",
            "99.3 99.3\n",
            "Saving model here\n",
            "EPOCH: 17\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.0478571355342865 Batch_id=468 Accuracy=99.18: 100%|██████████| 469/469 [00:11<00:00, 39.25it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0229, Accuracy: 9928/10000 (99.28%)\n",
            "\n",
            "99.3 99.28\n",
            "EPOCH: 18\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.17101234197616577 Batch_id=468 Accuracy=99.26: 100%|██████████| 469/469 [00:12<00:00, 37.98it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0228, Accuracy: 9925/10000 (99.25%)\n",
            "\n",
            "99.3 99.25\n",
            "EPOCH: 19\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.06769631057977676 Batch_id=468 Accuracy=99.16: 100%|██████████| 469/469 [00:12<00:00, 38.13it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0224, Accuracy: 9927/10000 (99.27%)\n",
            "\n",
            "99.3 99.27\n",
            "EPOCH: 20\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.012907758355140686 Batch_id=468 Accuracy=99.23: 100%|██████████| 469/469 [00:12<00:00, 38.26it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0228, Accuracy: 9925/10000 (99.25%)\n",
            "\n",
            "99.3 99.25\n",
            "EPOCH: 21\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.003455857513472438 Batch_id=468 Accuracy=99.15: 100%|██████████| 469/469 [00:12<00:00, 38.36it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0226, Accuracy: 9927/10000 (99.27%)\n",
            "\n",
            "99.3 99.27\n",
            "EPOCH: 22\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.004422960337251425 Batch_id=468 Accuracy=99.18: 100%|██████████| 469/469 [00:12<00:00, 37.94it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0222, Accuracy: 9927/10000 (99.27%)\n",
            "\n",
            "99.3 99.27\n",
            "EPOCH: 23\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.01312565803527832 Batch_id=468 Accuracy=99.17: 100%|██████████| 469/469 [00:12<00:00, 37.61it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0228, Accuracy: 9928/10000 (99.28%)\n",
            "\n",
            "99.3 99.28\n",
            "EPOCH: 24\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.004812197759747505 Batch_id=468 Accuracy=99.22: 100%|██████████| 469/469 [00:12<00:00, 38.05it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0222, Accuracy: 9926/10000 (99.26%)\n",
            "\n",
            "99.3 99.26\n",
            "EPOCH: 25\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.030738502740859985 Batch_id=468 Accuracy=99.26: 100%|██████████| 469/469 [00:12<00:00, 38.21it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0230, Accuracy: 9926/10000 (99.26%)\n",
            "\n",
            "99.3 99.26\n",
            "EPOCH: 26\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.018104668706655502 Batch_id=468 Accuracy=99.18: 100%|██████████| 469/469 [00:12<00:00, 38.60it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0220, Accuracy: 9928/10000 (99.28%)\n",
            "\n",
            "99.3 99.28\n",
            "EPOCH: 27\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.04104657843708992 Batch_id=468 Accuracy=99.24: 100%|██████████| 469/469 [00:12<00:00, 38.61it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0231, Accuracy: 9928/10000 (99.28%)\n",
            "\n",
            "99.3 99.28\n",
            "EPOCH: 28\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.002122580772265792 Batch_id=468 Accuracy=99.20: 100%|██████████| 469/469 [00:12<00:00, 38.38it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0222, Accuracy: 9929/10000 (99.29%)\n",
            "\n",
            "99.3 99.29\n",
            "EPOCH: 29\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.01464646216481924 Batch_id=468 Accuracy=99.20: 100%|██████████| 469/469 [00:12<00:00, 36.83it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0227, Accuracy: 9930/10000 (99.30%)\n",
            "\n",
            "99.3 99.3\n",
            "Saving model here\n",
            "EPOCH: 30\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.009402308613061905 Batch_id=468 Accuracy=99.18: 100%|██████████| 469/469 [00:12<00:00, 37.20it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0221, Accuracy: 9923/10000 (99.23%)\n",
            "\n",
            "99.3 99.23\n",
            "EPOCH: 31\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.033617518842220306 Batch_id=468 Accuracy=99.16: 100%|██████████| 469/469 [00:11<00:00, 39.75it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0223, Accuracy: 9929/10000 (99.29%)\n",
            "\n",
            "99.3 99.29\n",
            "EPOCH: 32\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.003985182847827673 Batch_id=468 Accuracy=99.19: 100%|██████████| 469/469 [00:12<00:00, 38.82it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0227, Accuracy: 9927/10000 (99.27%)\n",
            "\n",
            "99.3 99.27\n",
            "EPOCH: 33\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.023649848997592926 Batch_id=468 Accuracy=99.19: 100%|██████████| 469/469 [00:12<00:00, 38.97it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0226, Accuracy: 9926/10000 (99.26%)\n",
            "\n",
            "99.3 99.26\n",
            "EPOCH: 34\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.009338229894638062 Batch_id=468 Accuracy=99.24: 100%|██████████| 469/469 [00:12<00:00, 38.39it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0221, Accuracy: 9928/10000 (99.28%)\n",
            "\n",
            "99.3 99.28\n",
            "EPOCH: 35\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.0156644769012928 Batch_id=468 Accuracy=99.18: 100%|██████████| 469/469 [00:12<00:00, 38.12it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0230, Accuracy: 9927/10000 (99.27%)\n",
            "\n",
            "99.3 99.27\n",
            "EPOCH: 36\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.12693272531032562 Batch_id=468 Accuracy=99.26: 100%|██████████| 469/469 [00:12<00:00, 39.04it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0223, Accuracy: 9930/10000 (99.30%)\n",
            "\n",
            "99.3 99.3\n",
            "Saving model here\n",
            "EPOCH: 37\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.042733240872621536 Batch_id=468 Accuracy=99.17: 100%|██████████| 469/469 [00:12<00:00, 38.90it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0223, Accuracy: 9926/10000 (99.26%)\n",
            "\n",
            "99.3 99.26\n",
            "EPOCH: 38\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.10957914590835571 Batch_id=468 Accuracy=99.21: 100%|██████████| 469/469 [00:11<00:00, 39.39it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0225, Accuracy: 9926/10000 (99.26%)\n",
            "\n",
            "99.3 99.26\n",
            "EPOCH: 39\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.0208726916462183 Batch_id=468 Accuracy=99.22: 100%|██████████| 469/469 [00:12<00:00, 38.19it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0227, Accuracy: 9925/10000 (99.25%)\n",
            "\n",
            "99.3 99.25\n",
            "EPOCH: 40\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.04985083267092705 Batch_id=468 Accuracy=99.19: 100%|██████████| 469/469 [00:12<00:00, 37.75it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0228, Accuracy: 9927/10000 (99.27%)\n",
            "\n",
            "99.3 99.27\n",
            "EPOCH: 41\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.07496066391468048 Batch_id=468 Accuracy=99.25: 100%|██████████| 469/469 [00:12<00:00, 37.63it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0227, Accuracy: 9925/10000 (99.25%)\n",
            "\n",
            "99.3 99.25\n",
            "EPOCH: 42\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.019618330523371696 Batch_id=468 Accuracy=99.24: 100%|██████████| 469/469 [00:12<00:00, 38.24it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0225, Accuracy: 9927/10000 (99.27%)\n",
            "\n",
            "99.3 99.27\n",
            "EPOCH: 43\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.008292394690215588 Batch_id=468 Accuracy=99.20: 100%|██████████| 469/469 [00:12<00:00, 38.50it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0226, Accuracy: 9929/10000 (99.29%)\n",
            "\n",
            "99.3 99.29\n",
            "EPOCH: 44\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.04374505206942558 Batch_id=468 Accuracy=99.19: 100%|██████████| 469/469 [00:12<00:00, 39.05it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0221, Accuracy: 9927/10000 (99.27%)\n",
            "\n",
            "99.3 99.27\n",
            "EPOCH: 45\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.019316690042614937 Batch_id=468 Accuracy=99.17: 100%|██████████| 469/469 [00:12<00:00, 38.40it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0222, Accuracy: 9928/10000 (99.28%)\n",
            "\n",
            "99.3 99.28\n",
            "EPOCH: 46\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.06526999920606613 Batch_id=468 Accuracy=99.17: 100%|██████████| 469/469 [00:12<00:00, 38.35it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0230, Accuracy: 9929/10000 (99.29%)\n",
            "\n",
            "99.3 99.29\n",
            "EPOCH: 47\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.01877213642001152 Batch_id=468 Accuracy=99.20: 100%|██████████| 469/469 [00:12<00:00, 37.91it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0224, Accuracy: 9930/10000 (99.30%)\n",
            "\n",
            "99.3 99.3\n",
            "Saving model here\n",
            "EPOCH: 48\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.010773216374218464 Batch_id=468 Accuracy=99.21: 100%|██████████| 469/469 [00:12<00:00, 37.83it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0230, Accuracy: 9927/10000 (99.27%)\n",
            "\n",
            "99.3 99.27\n",
            "EPOCH: 49\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.10083002597093582 Batch_id=468 Accuracy=99.24: 100%|██████████| 469/469 [00:12<00:00, 37.76it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0220, Accuracy: 9927/10000 (99.27%)\n",
            "\n",
            "99.3 99.27\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 336
        },
        "id": "Pr3ztiDZ1dWk",
        "outputId": "094ef550-39f6-4409-982c-e396d64957cb"
      },
      "source": [
        "plot_curves('5.6K parameters with dw conv')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1080x360 with 2 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "oZbnF3SVBkbv"
      },
      "source": [
        "#### BASENET TRAINING"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "kofAhs6xBhVp"
      },
      "source": [
        "epochs = 50\n",
        "optimizer = optim.Adam(basenet.parameters(), lr = 0.01)\n",
        "scheduler = StepLR(optimizer, step_size=5, gamma=0.1)\n",
        "path = './quantization/pytorch_model/basenet25k.pth'\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "dlsxi8iGCIa9",
        "outputId": "48f765f5-35ba-4932-b4b1-a5580667fe09"
      },
      "source": [
        "execute_model(basenet, save_best_model=True)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "EPOCH: 0\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "\r  0%|          | 0/469 [00:00<?, ?it/s]/usr/local/lib/python3.7/dist-packages/torch/utils/data/dataloader.py:481: UserWarning: This DataLoader will create 4 worker processes in total. Our suggested max number of worker in current system is 2, which is smaller than what this DataLoader is going to create. Please be aware that excessive worker creation might get DataLoader running slow or even freeze, lower the worker number to avoid potential slowness/freeze if necessary.\n",
            "  cpuset_checked))\n",
            "Loss=0.00558740459382534 Batch_id=468 Accuracy=99.72: 100%|██████████| 469/469 [00:11<00:00, 42.58it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0227, Accuracy: 9929/10000 (99.29%)\n",
            "\n",
            "99.35 99.29\n",
            "EPOCH: 1\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.0005575611721724272 Batch_id=468 Accuracy=99.78: 100%|██████████| 469/469 [00:10<00:00, 43.27it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0228, Accuracy: 9929/10000 (99.29%)\n",
            "\n",
            "99.35 99.29\n",
            "EPOCH: 2\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.008801541291177273 Batch_id=468 Accuracy=99.82: 100%|██████████| 469/469 [00:11<00:00, 42.39it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0219, Accuracy: 9933/10000 (99.33%)\n",
            "\n",
            "99.35 99.33\n",
            "EPOCH: 3\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.005467440467327833 Batch_id=468 Accuracy=99.84: 100%|██████████| 469/469 [00:11<00:00, 42.39it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0224, Accuracy: 9929/10000 (99.29%)\n",
            "\n",
            "99.35 99.29\n",
            "EPOCH: 4\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.02612931840121746 Batch_id=468 Accuracy=99.83: 100%|██████████| 469/469 [00:10<00:00, 42.84it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0222, Accuracy: 9932/10000 (99.32%)\n",
            "\n",
            "99.35 99.32\n",
            "EPOCH: 5\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.001837934018112719 Batch_id=468 Accuracy=99.85: 100%|██████████| 469/469 [00:10<00:00, 43.25it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0213, Accuracy: 9932/10000 (99.32%)\n",
            "\n",
            "99.35 99.32\n",
            "EPOCH: 6\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.014143336564302444 Batch_id=468 Accuracy=99.85: 100%|██████████| 469/469 [00:11<00:00, 42.52it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0216, Accuracy: 9932/10000 (99.32%)\n",
            "\n",
            "99.35 99.32\n",
            "EPOCH: 7\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.00477896723896265 Batch_id=468 Accuracy=99.86: 100%|██████████| 469/469 [00:11<00:00, 42.31it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0214, Accuracy: 9935/10000 (99.35%)\n",
            "\n",
            "99.35 99.35\n",
            "Saving model here\n",
            "EPOCH: 8\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.004245364107191563 Batch_id=468 Accuracy=99.87: 100%|██████████| 469/469 [00:11<00:00, 41.58it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0222, Accuracy: 9930/10000 (99.30%)\n",
            "\n",
            "99.35 99.3\n",
            "EPOCH: 9\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.0023852328304201365 Batch_id=468 Accuracy=99.88: 100%|██████████| 469/469 [00:11<00:00, 42.44it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0216, Accuracy: 9932/10000 (99.32%)\n",
            "\n",
            "99.35 99.32\n",
            "EPOCH: 10\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.023078732192516327 Batch_id=468 Accuracy=99.87: 100%|██████████| 469/469 [00:10<00:00, 43.02it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0218, Accuracy: 9931/10000 (99.31%)\n",
            "\n",
            "99.35 99.31\n",
            "EPOCH: 11\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.001029090373776853 Batch_id=468 Accuracy=99.86: 100%|██████████| 469/469 [00:10<00:00, 42.87it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0216, Accuracy: 9932/10000 (99.32%)\n",
            "\n",
            "99.35 99.32\n",
            "EPOCH: 12\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.0019513815641403198 Batch_id=468 Accuracy=99.85: 100%|██████████| 469/469 [00:10<00:00, 43.02it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0220, Accuracy: 9931/10000 (99.31%)\n",
            "\n",
            "99.35 99.31\n",
            "EPOCH: 13\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.0014543962897732854 Batch_id=468 Accuracy=99.87: 100%|██████████| 469/469 [00:10<00:00, 42.96it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0213, Accuracy: 9934/10000 (99.34%)\n",
            "\n",
            "99.35 99.34\n",
            "EPOCH: 14\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.005384864751249552 Batch_id=468 Accuracy=99.86: 100%|██████████| 469/469 [00:11<00:00, 41.81it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0213, Accuracy: 9934/10000 (99.34%)\n",
            "\n",
            "99.35 99.34\n",
            "EPOCH: 15\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.0007527198176831007 Batch_id=468 Accuracy=99.88: 100%|██████████| 469/469 [00:11<00:00, 41.40it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0221, Accuracy: 9931/10000 (99.31%)\n",
            "\n",
            "99.35 99.31\n",
            "EPOCH: 16\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.0005670698010362685 Batch_id=468 Accuracy=99.88: 100%|██████████| 469/469 [00:11<00:00, 42.33it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0216, Accuracy: 9931/10000 (99.31%)\n",
            "\n",
            "99.35 99.31\n",
            "EPOCH: 17\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.011619067750871181 Batch_id=468 Accuracy=99.85: 100%|██████████| 469/469 [00:11<00:00, 41.93it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0216, Accuracy: 9931/10000 (99.31%)\n",
            "\n",
            "99.35 99.31\n",
            "EPOCH: 18\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.010014846920967102 Batch_id=468 Accuracy=99.87: 100%|██████████| 469/469 [00:11<00:00, 41.18it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0218, Accuracy: 9933/10000 (99.33%)\n",
            "\n",
            "99.35 99.33\n",
            "EPOCH: 19\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.0011635733535513282 Batch_id=468 Accuracy=99.87: 100%|██████████| 469/469 [00:11<00:00, 41.99it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0212, Accuracy: 9934/10000 (99.34%)\n",
            "\n",
            "99.35 99.34\n",
            "EPOCH: 20\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.02658594399690628 Batch_id=468 Accuracy=99.86: 100%|██████████| 469/469 [00:11<00:00, 41.25it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0222, Accuracy: 9931/10000 (99.31%)\n",
            "\n",
            "99.35 99.31\n",
            "EPOCH: 21\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.012752990238368511 Batch_id=468 Accuracy=99.85: 100%|██████████| 469/469 [00:11<00:00, 41.11it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0214, Accuracy: 9933/10000 (99.33%)\n",
            "\n",
            "99.35 99.33\n",
            "EPOCH: 22\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.006269669160246849 Batch_id=468 Accuracy=99.86: 100%|██████████| 469/469 [00:11<00:00, 41.75it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0217, Accuracy: 9930/10000 (99.30%)\n",
            "\n",
            "99.35 99.3\n",
            "EPOCH: 23\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.00241866591386497 Batch_id=468 Accuracy=99.87: 100%|██████████| 469/469 [00:11<00:00, 41.17it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0213, Accuracy: 9933/10000 (99.33%)\n",
            "\n",
            "99.35 99.33\n",
            "EPOCH: 24\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.0014789925189688802 Batch_id=468 Accuracy=99.87: 100%|██████████| 469/469 [00:11<00:00, 41.96it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0217, Accuracy: 9931/10000 (99.31%)\n",
            "\n",
            "99.35 99.31\n",
            "EPOCH: 25\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.03143145143985748 Batch_id=468 Accuracy=99.86: 100%|██████████| 469/469 [00:11<00:00, 41.34it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0220, Accuracy: 9930/10000 (99.30%)\n",
            "\n",
            "99.35 99.3\n",
            "EPOCH: 26\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.004307493567466736 Batch_id=468 Accuracy=99.89: 100%|██████████| 469/469 [00:11<00:00, 41.55it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0213, Accuracy: 9933/10000 (99.33%)\n",
            "\n",
            "99.35 99.33\n",
            "EPOCH: 27\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.0007890099077485502 Batch_id=468 Accuracy=99.87: 100%|██████████| 469/469 [00:11<00:00, 41.41it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0210, Accuracy: 9933/10000 (99.33%)\n",
            "\n",
            "99.35 99.33\n",
            "EPOCH: 28\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.0028155960608273745 Batch_id=468 Accuracy=99.86: 100%|██████████| 469/469 [00:11<00:00, 40.56it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0211, Accuracy: 9933/10000 (99.33%)\n",
            "\n",
            "99.35 99.33\n",
            "EPOCH: 29\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.0015522785251960158 Batch_id=468 Accuracy=99.86: 100%|██████████| 469/469 [00:11<00:00, 40.87it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0221, Accuracy: 9929/10000 (99.29%)\n",
            "\n",
            "99.35 99.29\n",
            "EPOCH: 30\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.0015475740656256676 Batch_id=468 Accuracy=99.87: 100%|██████████| 469/469 [00:11<00:00, 41.71it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0215, Accuracy: 9931/10000 (99.31%)\n",
            "\n",
            "99.35 99.31\n",
            "EPOCH: 31\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.0021144957281649113 Batch_id=468 Accuracy=99.86: 100%|██████████| 469/469 [00:11<00:00, 41.19it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0214, Accuracy: 9931/10000 (99.31%)\n",
            "\n",
            "99.35 99.31\n",
            "EPOCH: 32\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.0050553795881569386 Batch_id=468 Accuracy=99.87: 100%|██████████| 469/469 [00:11<00:00, 40.97it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0223, Accuracy: 9930/10000 (99.30%)\n",
            "\n",
            "99.35 99.3\n",
            "EPOCH: 33\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.004191295709460974 Batch_id=468 Accuracy=99.84: 100%|██████████| 469/469 [00:11<00:00, 41.04it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0218, Accuracy: 9934/10000 (99.34%)\n",
            "\n",
            "99.35 99.34\n",
            "EPOCH: 34\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.010423672385513783 Batch_id=468 Accuracy=99.88: 100%|██████████| 469/469 [00:11<00:00, 40.53it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0207, Accuracy: 9935/10000 (99.35%)\n",
            "\n",
            "99.35 99.35\n",
            "Saving model here\n",
            "EPOCH: 35\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.002116902032867074 Batch_id=468 Accuracy=99.88: 100%|██████████| 469/469 [00:11<00:00, 41.59it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0223, Accuracy: 9930/10000 (99.30%)\n",
            "\n",
            "99.35 99.3\n",
            "EPOCH: 36\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.003453887300565839 Batch_id=468 Accuracy=99.89: 100%|██████████| 469/469 [00:11<00:00, 40.67it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0232, Accuracy: 9927/10000 (99.27%)\n",
            "\n",
            "99.35 99.27\n",
            "EPOCH: 37\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.0029388591647148132 Batch_id=468 Accuracy=99.88: 100%|██████████| 469/469 [00:11<00:00, 41.27it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0213, Accuracy: 9935/10000 (99.35%)\n",
            "\n",
            "99.35 99.35\n",
            "Saving model here\n",
            "EPOCH: 38\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.005716720595955849 Batch_id=468 Accuracy=99.89: 100%|██████████| 469/469 [00:11<00:00, 40.94it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0219, Accuracy: 9932/10000 (99.32%)\n",
            "\n",
            "99.35 99.32\n",
            "EPOCH: 39\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.0008876295760273933 Batch_id=468 Accuracy=99.86: 100%|██████████| 469/469 [00:11<00:00, 41.05it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0217, Accuracy: 9932/10000 (99.32%)\n",
            "\n",
            "99.35 99.32\n",
            "EPOCH: 40\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.0008701427723281085 Batch_id=468 Accuracy=99.88: 100%|██████████| 469/469 [00:11<00:00, 40.36it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0216, Accuracy: 9932/10000 (99.32%)\n",
            "\n",
            "99.35 99.32\n",
            "EPOCH: 41\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.009804162196815014 Batch_id=468 Accuracy=99.86: 100%|██████████| 469/469 [00:11<00:00, 40.55it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0212, Accuracy: 9932/10000 (99.32%)\n",
            "\n",
            "99.35 99.32\n",
            "EPOCH: 42\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.02477221190929413 Batch_id=468 Accuracy=99.87: 100%|██████████| 469/469 [00:12<00:00, 39.05it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0221, Accuracy: 9930/10000 (99.30%)\n",
            "\n",
            "99.35 99.3\n",
            "EPOCH: 43\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.0011001399252563715 Batch_id=468 Accuracy=99.86: 100%|██████████| 469/469 [00:12<00:00, 38.58it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0219, Accuracy: 9931/10000 (99.31%)\n",
            "\n",
            "99.35 99.31\n",
            "EPOCH: 44\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.003773842239752412 Batch_id=468 Accuracy=99.84: 100%|██████████| 469/469 [00:11<00:00, 41.25it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0213, Accuracy: 9933/10000 (99.33%)\n",
            "\n",
            "99.35 99.33\n",
            "EPOCH: 45\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.0025346192996948957 Batch_id=468 Accuracy=99.87: 100%|██████████| 469/469 [00:11<00:00, 40.21it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0218, Accuracy: 9931/10000 (99.31%)\n",
            "\n",
            "99.35 99.31\n",
            "EPOCH: 46\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.004092139657586813 Batch_id=468 Accuracy=99.85: 100%|██████████| 469/469 [00:11<00:00, 40.44it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0215, Accuracy: 9932/10000 (99.32%)\n",
            "\n",
            "99.35 99.32\n",
            "EPOCH: 47\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.0013715067179873586 Batch_id=468 Accuracy=99.87: 100%|██████████| 469/469 [00:11<00:00, 41.15it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0216, Accuracy: 9930/10000 (99.30%)\n",
            "\n",
            "99.35 99.3\n",
            "EPOCH: 48\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.007209875155240297 Batch_id=468 Accuracy=99.86: 100%|██████████| 469/469 [00:11<00:00, 41.02it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0224, Accuracy: 9931/10000 (99.31%)\n",
            "\n",
            "99.35 99.31\n",
            "EPOCH: 49\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.0020154661033302546 Batch_id=468 Accuracy=99.87: 100%|██████████| 469/469 [00:11<00:00, 41.49it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0216, Accuracy: 9932/10000 (99.32%)\n",
            "\n",
            "99.35 99.32\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 336
        },
        "id": "phqzeElBCL7j",
        "outputId": "b73bde24-fcc2-461e-aad7-b066290606e8"
      },
      "source": [
        "plot_curves('Basenet 25K')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1080x360 with 2 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "xsa5kqk4QtXM"
      },
      "source": [
        "#### 1.5K ONE CYCLE LR"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "nq5MKVXyQsu7"
      },
      "source": [
        "epochs = 50\n",
        "optimizer = optim.Adam(m.parameters(), lr = 0.01)\n",
        "scheduler = torch.optim.lr_scheduler.OneCycleLR(optimizer, max_lr=0.5, steps_per_epoch=len(train_loader), epochs=epochs)\n",
        "path = './quantization/pytorch_model/1.5k_one_cycle.pth'"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "jyZNli-4Rk_V",
        "outputId": "48ce9b2a-35d6-484d-f296-d395d2a625a0"
      },
      "source": [
        "execute_model(m, save_best_model=True)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "EPOCH: 0\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "  0%|          | 0/469 [00:00<?, ?it/s]/usr/local/lib/python3.7/dist-packages/torch/utils/data/dataloader.py:481: UserWarning: This DataLoader will create 4 worker processes in total. Our suggested max number of worker in current system is 2, which is smaller than what this DataLoader is going to create. Please be aware that excessive worker creation might get DataLoader running slow or even freeze, lower the worker number to avoid potential slowness/freeze if necessary.\n",
            "  cpuset_checked))\n",
            "Loss=0.09385713934898376 Batch_id=468 Accuracy=95.64: 100%|██████████| 469/469 [00:18<00:00, 25.92it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.1047, Accuracy: 9674/10000 (96.74%)\n",
            "\n",
            "96.74 96.74\n",
            "Saving model here\n",
            "EPOCH: 1\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.10431339591741562 Batch_id=468 Accuracy=96.82: 100%|██████████| 469/469 [00:17<00:00, 26.17it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0696, Accuracy: 9765/10000 (97.65%)\n",
            "\n",
            "97.65 97.65\n",
            "Saving model here\n",
            "EPOCH: 2\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.029705272987484932 Batch_id=468 Accuracy=97.08: 100%|██████████| 469/469 [00:18<00:00, 25.61it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0827, Accuracy: 9746/10000 (97.46%)\n",
            "\n",
            "97.65 97.46\n",
            "EPOCH: 3\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.09929722547531128 Batch_id=468 Accuracy=97.39: 100%|██████████| 469/469 [00:18<00:00, 25.88it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0633, Accuracy: 9791/10000 (97.91%)\n",
            "\n",
            "97.91 97.91\n",
            "Saving model here\n",
            "EPOCH: 4\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.03752538934350014 Batch_id=468 Accuracy=97.50: 100%|██████████| 469/469 [00:18<00:00, 25.76it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0963, Accuracy: 9704/10000 (97.04%)\n",
            "\n",
            "97.91 97.04\n",
            "EPOCH: 5\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.03874589130282402 Batch_id=468 Accuracy=97.58: 100%|██████████| 469/469 [00:18<00:00, 25.70it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0642, Accuracy: 9803/10000 (98.03%)\n",
            "\n",
            "98.03 98.03\n",
            "Saving model here\n",
            "EPOCH: 6\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.05910563841462135 Batch_id=468 Accuracy=97.69: 100%|██████████| 469/469 [00:18<00:00, 25.88it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0549, Accuracy: 9826/10000 (98.26%)\n",
            "\n",
            "98.26 98.26\n",
            "Saving model here\n",
            "EPOCH: 7\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.06552813202142715 Batch_id=468 Accuracy=97.85: 100%|██████████| 469/469 [00:18<00:00, 25.74it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0643, Accuracy: 9814/10000 (98.14%)\n",
            "\n",
            "98.26 98.14\n",
            "EPOCH: 8\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.003482641652226448 Batch_id=468 Accuracy=97.84: 100%|██████████| 469/469 [00:18<00:00, 25.91it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0748, Accuracy: 9757/10000 (97.57%)\n",
            "\n",
            "98.26 97.57\n",
            "EPOCH: 9\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.057568471878767014 Batch_id=468 Accuracy=97.92: 100%|██████████| 469/469 [00:19<00:00, 24.68it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0584, Accuracy: 9817/10000 (98.17%)\n",
            "\n",
            "98.26 98.17\n",
            "EPOCH: 10\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.04158405959606171 Batch_id=468 Accuracy=98.07: 100%|██████████| 469/469 [00:19<00:00, 24.55it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0506, Accuracy: 9843/10000 (98.43%)\n",
            "\n",
            "98.43 98.43\n",
            "Saving model here\n",
            "EPOCH: 11\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.044068530201911926 Batch_id=468 Accuracy=98.08: 100%|██████████| 469/469 [00:19<00:00, 23.87it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0497, Accuracy: 9843/10000 (98.43%)\n",
            "\n",
            "98.43 98.43\n",
            "Saving model here\n",
            "EPOCH: 12\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.02702931873500347 Batch_id=468 Accuracy=97.98: 100%|██████████| 469/469 [00:19<00:00, 24.29it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0449, Accuracy: 9854/10000 (98.54%)\n",
            "\n",
            "98.54 98.54\n",
            "Saving model here\n",
            "EPOCH: 13\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.19608068466186523 Batch_id=468 Accuracy=98.14: 100%|██████████| 469/469 [00:18<00:00, 25.08it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0468, Accuracy: 9857/10000 (98.57%)\n",
            "\n",
            "98.57 98.57\n",
            "Saving model here\n",
            "EPOCH: 14\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.09502141922712326 Batch_id=468 Accuracy=98.10: 100%|██████████| 469/469 [00:19<00:00, 24.06it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0459, Accuracy: 9860/10000 (98.60%)\n",
            "\n",
            "98.6 98.6\n",
            "Saving model here\n",
            "EPOCH: 15\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.1847274750471115 Batch_id=468 Accuracy=98.14: 100%|██████████| 469/469 [00:18<00:00, 24.74it/s]\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\n",
            "Test set: Average loss: 0.0528, Accuracy: 9830/10000 (98.30%)\n",
            "\n",
            "98.6 98.3\n",
            "EPOCH: 16\n"
          ]
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "Loss=0.017349306493997574 Batch_id=468 Accuracy=98.17: 100%|██████████| 469/469 [00:18<00:00, 24.88it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0509, Accuracy: 9837/10000 (98.37%)\n",
            "\n",
            "98.6 98.37\n",
            "EPOCH: 17\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.06058394908905029 Batch_id=468 Accuracy=98.27: 100%|██████████| 469/469 [00:19<00:00, 24.32it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0465, Accuracy: 9854/10000 (98.54%)\n",
            "\n",
            "98.6 98.54\n",
            "EPOCH: 18\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.04139604791998863 Batch_id=468 Accuracy=98.19: 100%|██████████| 469/469 [00:19<00:00, 24.13it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0458, Accuracy: 9852/10000 (98.52%)\n",
            "\n",
            "98.6 98.52\n",
            "EPOCH: 19\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.06869810074567795 Batch_id=468 Accuracy=98.25: 100%|██████████| 469/469 [00:19<00:00, 23.66it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0452, Accuracy: 9855/10000 (98.55%)\n",
            "\n",
            "98.6 98.55\n",
            "EPOCH: 20\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.15161560475826263 Batch_id=468 Accuracy=98.30: 100%|██████████| 469/469 [00:19<00:00, 24.46it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0434, Accuracy: 9860/10000 (98.60%)\n",
            "\n",
            "98.6 98.6\n",
            "Saving model here\n",
            "EPOCH: 21\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.02581685222685337 Batch_id=468 Accuracy=98.17: 100%|██████████| 469/469 [00:19<00:00, 24.02it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0594, Accuracy: 9819/10000 (98.19%)\n",
            "\n",
            "98.6 98.19\n",
            "EPOCH: 22\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.06287331134080887 Batch_id=468 Accuracy=98.34: 100%|██████████| 469/469 [00:19<00:00, 23.92it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0390, Accuracy: 9873/10000 (98.73%)\n",
            "\n",
            "98.73 98.73\n",
            "Saving model here\n",
            "EPOCH: 23\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.010902014560997486 Batch_id=468 Accuracy=98.29: 100%|██████████| 469/469 [00:19<00:00, 24.33it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0464, Accuracy: 9853/10000 (98.53%)\n",
            "\n",
            "98.73 98.53\n",
            "EPOCH: 24\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.040428802371025085 Batch_id=468 Accuracy=98.30: 100%|██████████| 469/469 [00:18<00:00, 24.86it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0381, Accuracy: 9872/10000 (98.72%)\n",
            "\n",
            "98.73 98.72\n",
            "EPOCH: 25\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.03805513679981232 Batch_id=468 Accuracy=98.39: 100%|██████████| 469/469 [00:19<00:00, 24.32it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0547, Accuracy: 9831/10000 (98.31%)\n",
            "\n",
            "98.73 98.31\n",
            "EPOCH: 26\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.02663888782262802 Batch_id=468 Accuracy=98.36: 100%|██████████| 469/469 [00:19<00:00, 24.52it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0469, Accuracy: 9857/10000 (98.57%)\n",
            "\n",
            "98.73 98.57\n",
            "EPOCH: 27\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.024576576426625252 Batch_id=468 Accuracy=98.36: 100%|██████████| 469/469 [00:18<00:00, 24.71it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0428, Accuracy: 9869/10000 (98.69%)\n",
            "\n",
            "98.73 98.69\n",
            "EPOCH: 28\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.08774302154779434 Batch_id=468 Accuracy=98.39: 100%|██████████| 469/469 [00:18<00:00, 24.90it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0408, Accuracy: 9871/10000 (98.71%)\n",
            "\n",
            "98.73 98.71\n",
            "EPOCH: 29\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.02231641672551632 Batch_id=468 Accuracy=98.41: 100%|██████████| 469/469 [00:18<00:00, 25.17it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0508, Accuracy: 9839/10000 (98.39%)\n",
            "\n",
            "98.73 98.39\n",
            "EPOCH: 30\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.054371997714042664 Batch_id=468 Accuracy=98.40: 100%|██████████| 469/469 [00:18<00:00, 24.94it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0366, Accuracy: 9881/10000 (98.81%)\n",
            "\n",
            "98.81 98.81\n",
            "Saving model here\n",
            "EPOCH: 31\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.05387061461806297 Batch_id=468 Accuracy=98.44: 100%|██████████| 469/469 [00:19<00:00, 24.58it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0455, Accuracy: 9862/10000 (98.62%)\n",
            "\n",
            "98.81 98.62\n",
            "EPOCH: 32\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.06623388081789017 Batch_id=468 Accuracy=98.43: 100%|██████████| 469/469 [00:18<00:00, 24.93it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0362, Accuracy: 9892/10000 (98.92%)\n",
            "\n",
            "98.92 98.92\n",
            "Saving model here\n",
            "EPOCH: 33\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.00817026849836111 Batch_id=468 Accuracy=98.44: 100%|██████████| 469/469 [00:18<00:00, 25.17it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0357, Accuracy: 9891/10000 (98.91%)\n",
            "\n",
            "98.92 98.91\n",
            "EPOCH: 34\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.016855401918292046 Batch_id=468 Accuracy=98.45: 100%|██████████| 469/469 [00:18<00:00, 25.06it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0497, Accuracy: 9848/10000 (98.48%)\n",
            "\n",
            "98.92 98.48\n",
            "EPOCH: 35\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.004988791886717081 Batch_id=468 Accuracy=98.47: 100%|██████████| 469/469 [00:18<00:00, 25.47it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0429, Accuracy: 9871/10000 (98.71%)\n",
            "\n",
            "98.92 98.71\n",
            "EPOCH: 36\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.0026717770379036665 Batch_id=468 Accuracy=98.46: 100%|██████████| 469/469 [00:18<00:00, 24.86it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0419, Accuracy: 9866/10000 (98.66%)\n",
            "\n",
            "98.92 98.66\n",
            "EPOCH: 37\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.035886988043785095 Batch_id=468 Accuracy=98.46: 100%|██████████| 469/469 [00:18<00:00, 25.03it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0461, Accuracy: 9858/10000 (98.58%)\n",
            "\n",
            "98.92 98.58\n",
            "EPOCH: 38\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.1082523986697197 Batch_id=468 Accuracy=98.44: 100%|██████████| 469/469 [00:18<00:00, 24.74it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0427, Accuracy: 9866/10000 (98.66%)\n",
            "\n",
            "98.92 98.66\n",
            "EPOCH: 39\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.08327525854110718 Batch_id=468 Accuracy=98.49: 100%|██████████| 469/469 [00:18<00:00, 25.82it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0425, Accuracy: 9875/10000 (98.75%)\n",
            "\n",
            "98.92 98.75\n",
            "EPOCH: 40\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.01462333183735609 Batch_id=468 Accuracy=98.45: 100%|██████████| 469/469 [00:18<00:00, 25.59it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0462, Accuracy: 9865/10000 (98.65%)\n",
            "\n",
            "98.92 98.65\n",
            "EPOCH: 41\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.03411365672945976 Batch_id=468 Accuracy=98.56: 100%|██████████| 469/469 [00:18<00:00, 24.87it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0472, Accuracy: 9853/10000 (98.53%)\n",
            "\n",
            "98.92 98.53\n",
            "EPOCH: 42\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.1219724714756012 Batch_id=468 Accuracy=98.49: 100%|██████████| 469/469 [00:19<00:00, 24.24it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0490, Accuracy: 9856/10000 (98.56%)\n",
            "\n",
            "98.92 98.56\n",
            "EPOCH: 43\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.039519596844911575 Batch_id=468 Accuracy=98.46: 100%|██████████| 469/469 [00:19<00:00, 24.46it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0478, Accuracy: 9839/10000 (98.39%)\n",
            "\n",
            "98.92 98.39\n",
            "EPOCH: 44\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.14146682620048523 Batch_id=468 Accuracy=98.53: 100%|██████████| 469/469 [00:19<00:00, 24.24it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0402, Accuracy: 9876/10000 (98.76%)\n",
            "\n",
            "98.92 98.76\n",
            "EPOCH: 45\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.0548822283744812 Batch_id=468 Accuracy=98.44: 100%|██████████| 469/469 [00:18<00:00, 24.71it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0373, Accuracy: 9886/10000 (98.86%)\n",
            "\n",
            "98.92 98.86\n",
            "EPOCH: 46\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.033068422228097916 Batch_id=468 Accuracy=98.48: 100%|██████████| 469/469 [00:18<00:00, 24.70it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0447, Accuracy: 9867/10000 (98.67%)\n",
            "\n",
            "98.92 98.67\n",
            "EPOCH: 47\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.03411846235394478 Batch_id=468 Accuracy=98.57: 100%|██████████| 469/469 [00:19<00:00, 24.60it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0479, Accuracy: 9854/10000 (98.54%)\n",
            "\n",
            "98.92 98.54\n",
            "EPOCH: 48\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.10830556601285934 Batch_id=468 Accuracy=98.52: 100%|██████████| 469/469 [00:19<00:00, 23.95it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0413, Accuracy: 9872/10000 (98.72%)\n",
            "\n",
            "98.92 98.72\n",
            "EPOCH: 49\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Loss=0.033116042613983154 Batch_id=468 Accuracy=98.50: 100%|██████████| 469/469 [00:19<00:00, 24.23it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0436, Accuracy: 9869/10000 (98.69%)\n",
            "\n",
            "98.92 98.69\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 336
        },
        "id": "T2aZsSdERl-q",
        "outputId": "135b6ea8-c8da-4e03-d793-0b6e4d2ca4c8"
      },
      "source": [
        "plot_curves('1.5k with one cycle lr')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1080x360 with 2 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "#### CIFARNET10"
      ],
      "metadata": {
        "id": "H82gEEwt8Jks"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "class CifarNet11(nn.Module):\n",
        "    def __init__(self):\n",
        "        super(CifarNet11, self).__init__()\n",
        "\n",
        "# FIRST MAJOR BLOCK\n",
        "        # first convolutional block\n",
        "        self.convblock1a = nn.Sequential(\n",
        "            nn.Conv2d(in_channels = 3,out_channels = 16, dilation  = 1,padding = 1, kernel_size= (3,3)),                     # in 32, out 32, RF 3\n",
        "            nn.ReLU(),\n",
        "            nn.BatchNorm2d(16)\n",
        "        )\n",
        "\n",
        "        self.convblock1b = nn.Sequential(\n",
        "            nn.Conv2d(in_channels = 16,out_channels = 16, dilation  = 1,padding = 1, kernel_size= (3,3)),                     # in 32, out 32, RF 3\n",
        "            nn.ReLU(),\n",
        "            nn.BatchNorm2d(16)\n",
        "        )\n",
        "        self.dropout1  = nn.Sequential(nn.Dropout(0.06))\n",
        "\n",
        "        # Using stride 2 without dilation instead of maxpooling\n",
        "        self.convpool1 = nn.MaxPool2d(2, 2)\n",
        "\n",
        "\n",
        "# SECOND MAJOR BLOCK\n",
        "          # first convolutional block\n",
        "        self.convblock2a = nn.Sequential(\n",
        "            # nn.Conv2d(in_channels = 16,out_channels = 32, dilation  = 1,padding = 1,kernel_size= (3,3)),                    # in 16, out 16, RF ?\n",
        "            nn.Conv2d(in_channels = 16,out_channels = 16,groups = 16, dilation  = 1,padding = 1,kernel_size= (3,3)),\n",
        "            nn.Conv2d(in_channels = 16,out_channels = 32, dilation = 1,padding = 0,kernel_size= (1,1)), # 8, 8, 3  \n",
        "            nn.ReLU(),\n",
        "            nn.BatchNorm2d(32)\n",
        "        ) # will the learning will be better in depth wise or in normal 3x3 convolutions?\n",
        "\n",
        "        self.convblock2b = nn.Sequential(\n",
        "            nn.Conv2d(in_channels = 32,out_channels = 32, dilation  = 1,padding = 1, kernel_size= (3,3)),                     # in 16, out 16, RF 3\n",
        "            nn.ReLU(),\n",
        "            nn.BatchNorm2d(32)\n",
        "        )\n",
        "        self.dropout2  = nn.Sequential(nn.Dropout(0.07))\n",
        "\n",
        "        # Using stride 2 without dilation to simulate max pooling\n",
        "        self.convpool2 = nn.MaxPool2d(2, 2)\n",
        "\n",
        "# THIRD MAJOR BLOCK\n",
        "         # first convolutional block\n",
        "        self.convblock3a = nn.Sequential(\n",
        "            # nn.Conv2d(in_channels = 32,out_channels = 64, dilation  = 1,padding = 1,kernel_size= (3,3)), \n",
        "            nn.Conv2d(in_channels = 32,out_channels = 32,groups = 32, dilation  = 1,padding = 1,kernel_size= (3,3)),        # in 8, out 8, RF ?\n",
        "            nn.Conv2d(in_channels = 32,out_channels = 64, dilation = 1,padding = 0,kernel_size= (1,1)), # 8, 8, 3           # in 8, out 8, RF ?\n",
        "            nn.ReLU(),\n",
        "            nn.BatchNorm2d(64)\n",
        "        )\n",
        "\n",
        "        self.convblock3b = nn.Sequential(\n",
        "            nn.Conv2d(in_channels = 64,out_channels = 64, dilation  = 2,padding = 2, kernel_size= (3,3)),                     # in 8, out 8, RF 3\n",
        "            nn.ReLU(),\n",
        "            nn.BatchNorm2d(64)\n",
        "        )\n",
        "        self.dropout3  = nn.Sequential(nn.Dropout(0.07))\n",
        "        # using dilation to simulate maxpooling\n",
        "        self.convpool3 = nn.MaxPool2d(2, 2)\n",
        "\n",
        "# FOURTH MAJOR BLOCK\n",
        "        # second convolutional block\n",
        "        self.convblock4a = nn.Sequential(\n",
        "            # nn.Conv2d(in_channels = 128,out_channels = 128, dilation  = 1,padding = 1,kernel_size= (3,3)), \n",
        "            nn.Conv2d(in_channels = 64,out_channels = 64, groups = 64, dilation  = 1,padding = 1,kernel_size= (3,3)),    # in 4, out 4, RF ?\n",
        "            nn.Conv2d(in_channels = 64,out_channels = 128, dilation  = 1,padding = 0,kernel_size= (1,1)),                  # in 4, out 4, RF ?\n",
        "            nn.ReLU(),\n",
        "            nn.BatchNorm2d(128)\n",
        "        )\n",
        "\n",
        "        self.convblock4a1 = nn.Sequential(\n",
        "            # nn.Conv2d(in_channels = 128,out_channels = 128, dilation  = 1,padding = 1,kernel_size= (3,3)), \n",
        "            nn.Conv2d(in_channels = 128,out_channels = 128, groups = 128, dilation  = 1,padding = 1,kernel_size= (3,3)),    # in 4, out 4,, RF ?\n",
        "            nn.Conv2d(in_channels = 128,out_channels = 192, dilation  = 1,padding = 0,kernel_size= (1,1)),                  # in 4, out 4, RF ?\n",
        "            nn.ReLU(),\n",
        "            nn.BatchNorm2d(192)\n",
        "        )\n",
        "\n",
        "        self.dropout4  = nn.Sequential(nn.Dropout(0.05))\n",
        "\n",
        "        self.convblock4b = nn.Sequential(\n",
        "            # nn.Conv2d(in_channels = 128,out_channels = 128, dilation  = 1,padding = 1,kernel_size= (3,3)), \n",
        "            nn.Conv2d(in_channels = 192,out_channels = 192, groups = 192, dilation  = 1,padding = 1,kernel_size= (3,3)),    # in 4, out 4,, RF ?\n",
        "            nn.Conv2d(in_channels = 192,out_channels = 260, dilation  = 1,padding = 0,kernel_size= (1,1)),                  # in 4, out 4, RF ?\n",
        "            nn.ReLU(),\n",
        "            nn.BatchNorm2d(260)\n",
        "        )\n",
        "\n",
        "        # self.convblock4b = nn.Sequential(\n",
        "        #     # nn.Conv2d(in_channels = 128,out_channels = 128, dilation  = 1,padding = 1,kernel_size= (3,3)), \n",
        "        #     nn.Conv2d(in_channels = 128,out_channels = 128, groups = 128, dilation  = 1,padding = 1,kernel_size= (3,3)),    # in 4, out 4,, RF ?\n",
        "        #     nn.Conv2d(in_channels = 128,out_channels = 256, dilation  = 1,padding = 0,kernel_size= (1,1)),                  # in 4, out 4, RF ?\n",
        "        #     nn.ReLU(),\n",
        "        #     nn.BatchNorm2d(256)\n",
        "        # )\n",
        "\n",
        "# GAP LAYER\n",
        "        self.gap = nn.AvgPool2d(4)                                                                                          # in 3, out 1, RF ?\n",
        "        self.convblockf = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=260, out_channels=10, kernel_size=(1, 1), padding=0, bias=False),\n",
        "        ) \n",
        "\n",
        "\n",
        "    def forward(self, x):\n",
        "        x = self.convblock1a(x)\n",
        "        x1 = self.convblock1b(x)\n",
        "        x = torch.add(x,x1)\n",
        "        x = self.dropout1(x)\n",
        "        x = self.convpool1(x)\n",
        "\n",
        "\n",
        "        x = self.convblock2a(x)\n",
        "        x1 = self.convblock2b(x)\n",
        "        x = torch.add(x,x1)\n",
        "        x = self.dropout2(x)\n",
        "        x = self.convpool2(x1)\n",
        "\n",
        "        x = self.convblock3a(x)\n",
        "        x1 = self.convblock3b(x)        \n",
        "        x = torch.add(x,x1)\n",
        "        x = self.dropout3(x)\n",
        "        x = self.convpool3(x1)\n",
        "\n",
        "        x = self.convblock4a(x)\n",
        "        x = self.convblock4a1(x)\n",
        "        x = self.dropout4(x)\n",
        "        x = self.convblock4b(x)\n",
        "        \n",
        "        \n",
        "        x = self.gap(x)\n",
        "        x = self.convblockf(x)\n",
        "        x = x.view(-1, 10)\n",
        "        return F.log_softmax(x, dim=-1)\n",
        "model = CifarNet11()\n"
      ],
      "metadata": {
        "id": "iBtFTyS58D2T"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "class CifarNet11a(nn.Module):\n",
        "    def __init__(self):\n",
        "        super(CifarNet11a, self).__init__()\n",
        "\n",
        "# FIRST MAJOR BLOCK\n",
        "        # first convolutional block\n",
        "        self.convblock1a = nn.Sequential(\n",
        "            nn.Conv2d(in_channels = 3,out_channels = 16, dilation  = 1,padding = 1, kernel_size= (3,3)),                     # in 32, out 32, RF 3\n",
        "            nn.ReLU(),\n",
        "            nn.BatchNorm2d(16)\n",
        "        )\n",
        "\n",
        "        self.convblock1b = nn.Sequential(\n",
        "            nn.Conv2d(in_channels = 16,out_channels = 16, dilation  = 1,padding = 1, kernel_size= (3,3)),                     # in 32, out 32, RF 3\n",
        "            nn.ReLU(),\n",
        "            nn.BatchNorm2d(16)\n",
        "        )\n",
        "        self.dropout1  = nn.Sequential(nn.Dropout(0.06))\n",
        "\n",
        "        # Using stride 2 without dilation instead of maxpooling\n",
        "        self.convpool1 = nn.MaxPool2d(2, 2)\n",
        "\n",
        "\n",
        "# SECOND MAJOR BLOCK\n",
        "          # first convolutional block\n",
        "        self.convblock2a = nn.Sequential(\n",
        "            nn.Conv2d(in_channels = 16,out_channels = 32, dilation  = 1,padding = 1,kernel_size= (3,3)),                    # in 16, out 16, RF ? \n",
        "            nn.ReLU(),\n",
        "            nn.BatchNorm2d(32)\n",
        "        ) # will the learning will be better in depth wise or in normal 3x3 convolutions?\n",
        "\n",
        "        self.convblock2b = nn.Sequential(\n",
        "            nn.Conv2d(in_channels = 32,out_channels = 32, dilation  = 1,padding = 1, kernel_size= (3,3)),                     # in 16, out 16, RF 3\n",
        "            nn.ReLU(),\n",
        "            nn.BatchNorm2d(32)\n",
        "        )\n",
        "        self.dropout2  = nn.Sequential(nn.Dropout(0.07))\n",
        "\n",
        "        # Using stride 2 without dilation to simulate max pooling\n",
        "        self.convpool2 = nn.MaxPool2d(2, 2)\n",
        "\n",
        "# THIRD MAJOR BLOCK\n",
        "         # first convolutional block\n",
        "        self.convblock3a = nn.Sequential(\n",
        "            nn.Conv2d(in_channels = 32,out_channels = 64, dilation  = 1,padding = 1,kernel_size= (3,3)), \n",
        "            nn.ReLU(),\n",
        "            nn.BatchNorm2d(64)\n",
        "        )\n",
        "\n",
        "        self.convblock3b = nn.Sequential(\n",
        "            nn.Conv2d(in_channels = 64,out_channels = 64, dilation  = 2,padding = 2, kernel_size= (3,3)),                     # in 8, out 8, RF 3\n",
        "            nn.ReLU(),\n",
        "            nn.BatchNorm2d(64)\n",
        "        )\n",
        "        self.dropout3  = nn.Sequential(nn.Dropout(0.07))\n",
        "        # using dilation to simulate maxpooling\n",
        "        self.convpool3 = nn.MaxPool2d(2, 2)\n",
        "\n",
        "# FOURTH MAJOR BLOCK\n",
        "        # second convolutional block\n",
        "        self.convblock4a = nn.Sequential(\n",
        "            nn.Conv2d(in_channels = 64,out_channels = 120, dilation  = 1,padding = 1,kernel_size= (3,3)), \n",
        "            nn.ReLU(),\n",
        "            nn.BatchNorm2d(120)\n",
        "        )\n",
        "# GAP LAYER\n",
        "        self.gap = nn.AvgPool2d(4)                                                                                          # in 3, out 1, RF ?\n",
        "        self.convblockf = nn.Sequential(\n",
        "            nn.Conv2d(in_channels=120, out_channels=10, kernel_size=(1, 1), padding=0, bias=False),\n",
        "        ) \n",
        "\n",
        "    def forward(self, x):\n",
        "        x = self.convblock1a(x)\n",
        "        x1 = self.convblock1b(x)\n",
        "        x = torch.add(x,x1)\n",
        "        x = self.dropout1(x)\n",
        "        x = self.convpool1(x)\n",
        "\n",
        "\n",
        "        x = self.convblock2a(x)\n",
        "        x1 = self.convblock2b(x)\n",
        "        x = torch.add(x,x1)\n",
        "        x = self.dropout2(x)\n",
        "        x = self.convpool2(x1)\n",
        "\n",
        "        x = self.convblock3a(x)\n",
        "        x1 = self.convblock3b(x)        \n",
        "        x = torch.add(x,x1)\n",
        "        x = self.dropout3(x)\n",
        "        x = self.convpool3(x1)\n",
        "\n",
        "        x = self.convblock4a(x)\n",
        "        x = self.gap(x)\n",
        "        x = self.convblockf(x)\n",
        "        x = x.view(-1, 10)\n",
        "        return F.log_softmax(x, dim=-1)\n",
        "model = CifarNet11a()"
      ],
      "metadata": {
        "id": "KT7pyqqGDyxi"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "UHX2b1nNckag"
      },
      "source": [
        "#### INFERENCE TIME EVALUATION"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "mSditzEtXTcD",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "46b08a3d-88cc-4460-9f85-801c60800418"
      },
      "source": [
        "path = './quantization/pytorch_model/normal_pytorch.pth'\n",
        "model = Net()\n",
        "model.load_state_dict(torch.load(path,map_location=torch.device('cpu')))  # Choose whatever GPU device number you want\n",
        "test1(model)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/torch/utils/data/dataloader.py:481: UserWarning: This DataLoader will create 4 worker processes in total. Our suggested max number of worker in current system is 2, which is smaller than what this DataLoader is going to create. Please be aware that excessive worker creation might get DataLoader running slow or even freeze, lower the worker number to avoid potential slowness/freeze if necessary.\n",
            "  cpuset_checked))\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0197, Accuracy: 9934/10000 (99.34%)\n",
            "\n",
            "time taken is 2.6732869148254395\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 461
        },
        "id": "4Umwrax2wDtB",
        "outputId": "409cbce8-104b-402b-ec77-4bb9f05846e8"
      },
      "source": [
        "path = './quantization/pytorch_model/mosaic_7k_dw.pth'\n",
        "model1 = Net1()\n",
        "model1.load_state_dict(torch.load(path, map_location=torch.device('cpu')))  # Choose whatever GPU device number you want\n",
        "test1(model1)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "error",
          "ename": "RuntimeError",
          "evalue": "ignored",
          "traceback": [
            "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
            "\u001b[0;31mRuntimeError\u001b[0m                              Traceback (most recent call last)",
            "\u001b[0;32m<ipython-input-18-03d24a1a1a94>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[0mpath\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m'./quantization/pytorch_model/mosaic_7k_dw.pth'\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      2\u001b[0m \u001b[0mmodel1\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mNet1\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 3\u001b[0;31m \u001b[0mmodel1\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mload_state_dict\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mtorch\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mload\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mpath\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mmap_location\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mtorch\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mdevice\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m'cpu'\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m  \u001b[0;31m# Choose whatever GPU device number you want\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      4\u001b[0m \u001b[0mtest1\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mmodel1\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py\u001b[0m in \u001b[0;36mload_state_dict\u001b[0;34m(self, state_dict, strict)\u001b[0m\n\u001b[1;32m   1481\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mlen\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0merror_msgs\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;34m>\u001b[0m \u001b[0;36m0\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1482\u001b[0m             raise RuntimeError('Error(s) in loading state_dict for {}:\\n\\t{}'.format(\n\u001b[0;32m-> 1483\u001b[0;31m                                self.__class__.__name__, \"\\n\\t\".join(error_msgs)))\n\u001b[0m\u001b[1;32m   1484\u001b[0m         \u001b[0;32mreturn\u001b[0m \u001b[0m_IncompatibleKeys\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mmissing_keys\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0munexpected_keys\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1485\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;31mRuntimeError\u001b[0m: Error(s) in loading state_dict for Net1:\n\tsize mismatch for convblock4.0.weight: copying a param with shape torch.Size([14, 10, 3, 3]) from checkpoint, the shape in current model is torch.Size([13, 10, 3, 3]).\n\tsize mismatch for convblock4.2.weight: copying a param with shape torch.Size([14]) from checkpoint, the shape in current model is torch.Size([13]).\n\tsize mismatch for convblock4.2.bias: copying a param with shape torch.Size([14]) from checkpoint, the shape in current model is torch.Size([13]).\n\tsize mismatch for convblock4.2.running_mean: copying a param with shape torch.Size([14]) from checkpoint, the shape in current model is torch.Size([13]).\n\tsize mismatch for convblock4.2.running_var: copying a param with shape torch.Size([14]) from checkpoint, the shape in current model is torch.Size([13]).\n\tsize mismatch for convblock5.0.weight: copying a param with shape torch.Size([16, 14, 3, 3]) from checkpoint, the shape in current model is torch.Size([16, 13, 3, 3])."
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "e9sChEdoRLs8",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "9fe4499e-03bb-47a8-88b1-4eb68090809f"
      },
      "source": [
        "path = './quantization/pytorch_model/mosaic_7k_dw_new.pth'\n",
        "model2 = Net1()\n",
        "model2.load_state_dict(torch.load(path,map_location=torch.device('cpu')))  # Choose whatever GPU device number you want\n",
        "test1(model2)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/torch/utils/data/dataloader.py:481: UserWarning: This DataLoader will create 4 worker processes in total. Our suggested max number of worker in current system is 2, which is smaller than what this DataLoader is going to create. Please be aware that excessive worker creation might get DataLoader running slow or even freeze, lower the worker number to avoid potential slowness/freeze if necessary.\n",
            "  cpuset_checked))\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0228, Accuracy: 9933/10000 (99.33%)\n",
            "\n",
            "time taken is 3.3572731018066406\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "axo7d1OUfUdy",
        "outputId": "56da6472-323e-4eac-fc95-cbcfbdcc3a37"
      },
      "source": [
        "path = './quantization/pytorch_model/smallest_pytorch.pth'\n",
        "smodel = SNet()\n",
        "smodel.load_state_dict(torch.load(path,map_location=torch.device('cpu')))  # Choose whatever GPU device number you want\n",
        "test1(smodel)\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/torch/utils/data/dataloader.py:481: UserWarning: This DataLoader will create 4 worker processes in total. Our suggested max number of worker in current system is 2, which is smaller than what this DataLoader is going to create. Please be aware that excessive worker creation might get DataLoader running slow or even freeze, lower the worker number to avoid potential slowness/freeze if necessary.\n",
            "  cpuset_checked))\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0484, Accuracy: 9835/10000 (98.35%)\n",
            "\n",
            "time taken is 2.521306037902832\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "yA9Y4HukQ-B0",
        "outputId": "65210b61-5ac5-481e-cf39-745c849ec596"
      },
      "source": [
        "path = './quantization/pytorch_model/small_pytorch.pth'\n",
        "cmodel = CNet()\n",
        "cmodel.load_state_dict(torch.load(path,map_location=torch.device('cpu')))  # Choose whatever GPU device number you want\n",
        "test1(cmodel)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/torch/utils/data/dataloader.py:481: UserWarning: This DataLoader will create 4 worker processes in total. Our suggested max number of worker in current system is 2, which is smaller than what this DataLoader is going to create. Please be aware that excessive worker creation might get DataLoader running slow or even freeze, lower the worker number to avoid potential slowness/freeze if necessary.\n",
            "  cpuset_checked))\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0224, Accuracy: 9927/10000 (99.27%)\n",
            "\n",
            "time taken is 2.762350559234619\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "wF6rk7ymuTYZ",
        "outputId": "17e9fa01-6dd4-4454-d41a-3dd831144ffd"
      },
      "source": [
        "path = './quantization/pytorch_model/mosaic_5K_dw_new.pth'\n",
        "cmodel = CNet1()\n",
        "cmodel.load_state_dict(torch.load(path,map_location=torch.device('cpu')))  # Choose whatever GPU device number you want\n",
        "test1(cmodel)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/torch/utils/data/dataloader.py:481: UserWarning: This DataLoader will create 4 worker processes in total. Our suggested max number of worker in current system is 2, which is smaller than what this DataLoader is going to create. Please be aware that excessive worker creation might get DataLoader running slow or even freeze, lower the worker number to avoid potential slowness/freeze if necessary.\n",
            "  cpuset_checked))\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0224, Accuracy: 9930/10000 (99.30%)\n",
            "\n",
            "time taken is 2.747142791748047\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "TL71RRyNDn9A",
        "outputId": "debe4e8b-e4cd-4c64-89b0-057beadf9d89"
      },
      "source": [
        "path = './quantization/pytorch_model/basenet25k.pth'\n",
        "cmodel = BaseNet()\n",
        "cmodel.load_state_dict(torch.load(path,map_location=torch.device('cpu')))  # Choose whatever GPU device number you want\n",
        "test1(cmodel)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/torch/utils/data/dataloader.py:481: UserWarning: This DataLoader will create 4 worker processes in total. Our suggested max number of worker in current system is 2, which is smaller than what this DataLoader is going to create. Please be aware that excessive worker creation might get DataLoader running slow or even freeze, lower the worker number to avoid potential slowness/freeze if necessary.\n",
            "  cpuset_checked))\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.0213, Accuracy: 9935/10000 (99.35%)\n",
            "\n",
            "time taken is 3.368082046508789\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "feDEe-J-JeI_",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "693a6891-c8cc-4849-ef82-98f572e0cd10"
      },
      "source": [
        "path = './mos_ml_cif_100/normal_pytorch_dw.pth'\n",
        "# cmodel = BaseNet()\n",
        "model.load_state_dict(torch.load(path,map_location=torch.device('cpu')))  # Choose whatever GPU device number you want\n",
        "test1(model)\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/torch/utils/data/dataloader.py:481: UserWarning: This DataLoader will create 4 worker processes in total. Our suggested max number of worker in current system is 2, which is smaller than what this DataLoader is going to create. Please be aware that excessive worker creation might get DataLoader running slow or even freeze, lower the worker number to avoid potential slowness/freeze if necessary.\n",
            "  cpuset_checked))\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.6639, Accuracy: 7990/10000 (79.90%)\n",
            "\n",
            "time taken is 8.8280930519104\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "path = './mos_ml_cif_100/normal_pytorch_wdw.pth'\n",
        "model.load_state_dict(torch.load(path,map_location=torch.device('cpu')))  # Choose whatever GPU device number you want\n",
        "test1(model)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "WUHMzRaZ8sV5",
        "outputId": "2a53dfdb-a387-435e-d2ed-b0f99b3c3fe8"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/torch/utils/data/dataloader.py:481: UserWarning: This DataLoader will create 4 worker processes in total. Our suggested max number of worker in current system is 2, which is smaller than what this DataLoader is going to create. Please be aware that excessive worker creation might get DataLoader running slow or even freeze, lower the worker number to avoid potential slowness/freeze if necessary.\n",
            "  cpuset_checked))\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Test set: Average loss: 0.6477, Accuracy: 8157/10000 (81.57%)\n",
            "\n",
            "time taken is 8.517145156860352\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        ""
      ],
      "metadata": {
        "id": "Vc8d4s0Y88Yy"
      },
      "execution_count": null,
      "outputs": []
    }
  ]
}