cifar10_trt.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "model_name = \"res50\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# -*- coding: utf-8 -*-\n",
    "'''Train CIFAR10 with PyTorch.'''\n",
    "\n",
    "from __future__ import print_function\n",
    "\n",
    "import torch\n",
    "import torch.nn as nn\n",
    "import torch.optim as optim\n",
    "import torch.nn.functional as F\n",
    "import torch.backends.cudnn as cudnn\n",
    "\n",
    "import torchvision\n",
    "import torchvision.transforms as transforms\n",
    "\n",
    "import os\n",
    "import argparse\n",
    "\n",
    "from models import *\n",
    "\n",
    "from torch2trt import torch2trt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "from torchvision.models import *"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "==> Preparing data..\n",
      "Files already downloaded and verified\n",
      "Files already downloaded and verified\n",
      "torch.Size([3, 32, 32])\n"
     ]
    }
   ],
   "source": [
    "# Data\n",
    "print('==> Preparing data..')\n",
    "transform_train = transforms.Compose([\n",
    "    transforms.RandomCrop(32, padding=4),\n",
    "    transforms.RandomHorizontalFlip(),\n",
    "    transforms.ToTensor(),\n",
    "    transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010)),\n",
    "])\n",
    "\n",
    "transform_test = transforms.Compose([\n",
    "    transforms.ToTensor(),\n",
    "    transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010)),\n",
    "])\n",
    "\n",
    "bs = 256\n",
    "\n",
    "trainset = torchvision.datasets.CIFAR10(root='./data', train=True, download=True, transform=transform_train)\n",
    "trainloader = torch.utils.data.DataLoader(trainset, batch_size=bs, shuffle=True, num_workers=8)\n",
    "\n",
    "testset = torchvision.datasets.CIFAR10(root='./data', train=False, download=True, transform=transform_test)\n",
    "testloader = torch.utils.data.DataLoader(testset, batch_size=100, shuffle=False, num_workers=8)\n",
    "\n",
    "print(trainset[0][0].shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([101, 3, 32, 32])"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "x = trainset[0][0].unsqueeze(0)\n",
    "for i in range(100):\n",
    "    x = torch.cat((x, trainset[i][0].unsqueeze(0)))\n",
    "x.size()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# make model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "class mymodel(nn.Module):\n",
    "    def __init__(self, basemodeel):\n",
    "        super(mymodel, self).__init__()\n",
    "        self.features = basemodel\n",
    "        self.conv1 = nn.Conv2d(3, 64, kernel_size=3, padding=1, bias=False)\n",
    "        if model_name == \"res34\" or model_name == \"res18\":\n",
    "            num_ch = 512\n",
    "        else:\n",
    "            num_ch = 2048\n",
    "        \n",
    "        self.avgpool = nn.AdaptiveAvgPool2d(1)\n",
    "        self.fc1 = nn.Conv2d(num_ch, 10, 1)\n",
    "        \n",
    "    def forward(self, x):\n",
    "        x = self.conv1(x)\n",
    "        x = self.features(x)\n",
    "        x = self.avgpool(x)#.squeeze(2).squeeze(2)\n",
    "        x = self.fc1(x)\n",
    "        return x"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([1, 10, 1, 1])"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "if model_name == \"res18\":\n",
    "    basemodel = resnet18(pretrained=True)\n",
    "elif model_name == \"res34\":\n",
    "    basemodel = resnet34(pretrained=True)\n",
    "elif model_name == \"res50\":\n",
    "    basemodel = resnet50(pretrained=True)\n",
    "    \n",
    "basemodel = nn.Sequential(*list(basemodel.children())[1:-2])\n",
    "net = mymodel(basemodel) \n",
    "\n",
    "inp = torch.rand(1,3,32,32)\n",
    "net(inp).size()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "net = net.to(\"cuda\")\n",
    "device = \"cuda\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "# load checkpoint\n",
    "checkpoint = torch.load('./checkpoint/{}-ckpt.t7'.format(model_name))\n",
    "net.load_state_dict(checkpoint['net'])\n",
    "best_acc = checkpoint['acc']\n",
    "start_epoch = checkpoint['epoch']\n",
    "\n",
    "criterion = nn.CrossEntropyLoss()\n",
    "optimizer = optim.SGD(net.parameters(), lr=1e-3, momentum=0.9, weight_decay=1e-4)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "# test eval\n",
    "def test(epoch, net):\n",
    "    global best_acc\n",
    "    net.eval()\n",
    "    test_loss = 0\n",
    "    correct = 0\n",
    "    total = 0\n",
    "    with torch.no_grad():\n",
    "        for batch_idx, (inputs, targets) in enumerate(testloader):\n",
    "            inputs, targets = inputs.to(device), targets.to(device)\n",
    "            outputs = net(inputs).squeeze(2).squeeze(2)\n",
    "            loss = criterion(outputs, targets)\n",
    "\n",
    "            test_loss += loss.item()\n",
    "            _, predicted = outputs.max(1)\n",
    "            total += targets.size(0)\n",
    "            correct += predicted.eq(targets).sum().item()\n",
    "\n",
    "    # Save checkpoint.\n",
    "    acc = 100.*correct/total\n",
    "    print(\"acc:\", acc)\n",
    "    return acc"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "acc: 88.89\n"
     ]
    }
   ],
   "source": [
    "accbefore=test(0, net)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "acc: 88.89\n"
     ]
    }
   ],
   "source": [
    "# make FP32 TensorRT model\n",
    "# define input\n",
    "input_size = [1, 3, 32, 32]\n",
    "x = torch.zeros(input_size).cuda()\n",
    "\n",
    "# convert to tensorrt models\n",
    "net_trt = torch2trt(net, [x], max_batch_size=256)\n",
    "\n",
    "fp32acc = test(0, net_trt)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "# make FP16 TensorRT model\n",
    "# define input\n",
    "input_size = [1, 3, 32, 32]\n",
    "x = torch.zeros(input_size).cuda()\n",
    "\n",
    "# convert to tensorrt models\n",
    "net_trt = torch2trt(net, [x], fp16_mode=True, max_batch_size=256)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "acc: 88.81\n"
     ]
    }
   ],
   "source": [
    "fp16acc = test(0, net_trt)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "# make INT8 TensorRT model\n",
    "# define input\n",
    "input_size = [1, 3, 32, 32]\n",
    "x = trainset[0][0].unsqueeze(0)\n",
    "for i in range(1000):\n",
    "    x = torch.cat((x, trainset[i][0].unsqueeze(0)))\n",
    "x = x.cuda()\n",
    "\n",
    "# convert to tensorrt models\n",
    "net_trt = torch2trt(net, [x], fp16_mode=True, int8_mode=True, max_batch_size=256)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "acc: 88.83\n"
     ]
    }
   ],
   "source": [
    "int8acc = test(0, net_trt)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}