Mixed Precision - Filter Comparison.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "import torch.nn as nn\n",
    "import torch.functional as F\n",
    "torch.backends.cudnn.benchmark = True\n",
    "from functools import partial\n",
    "import functools\n",
    "\n",
    "x = torch.randn(1).cuda()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Train"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "def train(m, it, half, shape):\n",
    "    x = torch.randn(shape).cuda()\n",
    "    m = m.cuda()\n",
    "    \n",
    "    if half:\n",
    "        x = x.half()\n",
    "        m = m.half()\n",
    "        \n",
    "    for i in range(it):\n",
    "        out = m(x)\n",
    "        loss = out.sum()\n",
    "        loss.backward()\n",
    "        if hasattr(m, 'zero_grad'):\n",
    "            m.zero_grad()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "def upsample(shape, num_blocks=10):\n",
    "    block = lambda: [nn.Conv2d(shape[1], shape[1], kernel_size=3, stride=1, padding=1)]\n",
    "    layers = []\n",
    "    for i in range(num_blocks): layers.extend(block())\n",
    "    return nn.Sequential(*layers)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 512 Filters"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "shape = [64,512,64,64]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "m_upsample = upsample(shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 7.72 s, sys: 4.97 s, total: 12.7 s\n",
      "Wall time: 14.5 s\n"
     ]
    }
   ],
   "source": [
    "%time train(m_upsample, it=20, shape=shape, half=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Half"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "m_upsample = upsample(shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 7 s, sys: 4.19 s, total: 11.2 s\n",
      "Wall time: 13.2 s\n"
     ]
    }
   ],
   "source": [
    "%time train(m_upsample, it=20, shape=shape, half=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 8 Filters"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "shape = [16,8,512,512]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "m_upsample = upsample(shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 4.08 s, sys: 3.47 s, total: 7.55 s\n",
      "Wall time: 7.67 s\n"
     ]
    }
   ],
   "source": [
    "%time train(m_upsample, it=50, shape=shape, half=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Half - slower performance"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "m_upsample = upsample(shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 4.99 s, sys: 4.02 s, total: 9.01 s\n",
      "Wall time: 9.37 s\n"
     ]
    }
   ],
   "source": [
    "%time train(m_upsample, it=50, shape=shape, half=True)"
   ]
  }
 ],
 "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": "3.7.1"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}