PaddleSpeech/.notebook/WarmupLR.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "d6a0e098",
   "metadata": {},
   "outputs": [],
   "source": [
    "from typing import Union\n",
    "\n",
    "import torch\n",
    "from torch.optim.lr_scheduler import _LRScheduler\n",
    "\n",
    "from typeguard import check_argument_types\n",
    "\n",
    "\n",
    "class WarmupLR(_LRScheduler):\n",
    "    \"\"\"The WarmupLR scheduler\n",
    "    This scheduler is almost same as NoamLR Scheduler except for following\n",
    "    difference:\n",
    "    NoamLR:\n",
    "        lr = optimizer.lr * model_size ** -0.5\n",
    "             * min(step ** -0.5, step * warmup_step ** -1.5)\n",
    "    WarmupLR:\n",
    "        lr = optimizer.lr * warmup_step ** 0.5\n",
    "             * min(step ** -0.5, step * warmup_step ** -1.5)\n",
    "    Note that the maximum lr equals to optimizer.lr in this scheduler.\n",
    "    \"\"\"\n",
    "\n",
    "    def __init__(\n",
    "        self,\n",
    "        optimizer: torch.optim.Optimizer,\n",
    "        warmup_steps: Union[int, float] = 25000,\n",
    "        last_epoch: int = -1,\n",
    "    ):\n",
    "        assert check_argument_types()\n",
    "        self.warmup_steps = warmup_steps\n",
    "\n",
    "        # __init__() must be invoked before setting field\n",
    "        # because step() is also invoked in __init__()\n",
    "        super().__init__(optimizer, last_epoch)\n",
    "\n",
    "    def __repr__(self):\n",
    "        return f\"{self.__class__.__name__}(warmup_steps={self.warmup_steps})\"\n",
    "\n",
    "    def get_lr(self):\n",
    "        step_num = self.last_epoch + 1\n",
    "        return [\n",
    "            lr\n",
    "            * self.warmup_steps ** 0.5\n",
    "            * min(step_num ** -0.5, step_num * self.warmup_steps ** -1.5)\n",
    "            for lr in self.base_lrs\n",
    "        ]\n",
    "\n",
    "    def set_step(self, step: int):\n",
    "        self.last_epoch = step"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "0d496677",
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch.optim as optim\n",
    "model = torch.nn.Linear(10, 200)\n",
    "optimizer = optim.Adam(model.parameters())\n",
    "scheduler = WarmupLR(optimizer, warmup_steps=25000)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "e3e3f3dc",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0 0.0 -1\n"
     ]
    }
   ],
   "source": [
    "infos = {}\n",
    "start_epoch = infos.get('epoch', -1) + 1\n",
    "cv_loss = infos.get('cv_loss', 0.0)\n",
    "step = infos.get('step', -1)\n",
    "print(start_epoch, cv_loss, step)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "dc3d550c",
   "metadata": {},
   "outputs": [],
   "source": [
    "scheduler.set_step(step)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "e527634e",
   "metadata": {},
   "outputs": [],
   "source": [
    "lrs=[]\n",
    "for i in range(100000):\n",
    "    scheduler.step()\n",
    "    lrs.append(scheduler.get_lr())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "f1452db9",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Collecting matplotlib\n",
      "  Downloading matplotlib-3.4.1-cp38-cp38-manylinux1_x86_64.whl (10.3 MB)\n",
      "\u001b[K     |████████████████████████████████| 10.3 MB 575 kB/s eta 0:00:01\n",
      "\u001b[?25hCollecting kiwisolver>=1.0.1\n",
      "  Downloading kiwisolver-1.3.1-cp38-cp38-manylinux1_x86_64.whl (1.2 MB)\n",
      "\u001b[K     |████████████████████████████████| 1.2 MB 465 kB/s eta 0:00:01\n",
      "\u001b[?25hRequirement already satisfied: pillow>=6.2.0 in /workspace/wenet/venv/lib/python3.8/site-packages (from matplotlib) (8.1.2)\n",
      "Requirement already satisfied: numpy>=1.16 in /workspace/wenet/venv/lib/python3.8/site-packages (from matplotlib) (1.20.1)\n",
      "Requirement already satisfied: python-dateutil>=2.7 in /workspace/wenet/venv/lib/python3.8/site-packages (from matplotlib) (2.8.1)\n",
      "Collecting cycler>=0.10\n",
      "  Downloading cycler-0.10.0-py2.py3-none-any.whl (6.5 kB)\n",
      "Requirement already satisfied: pyparsing>=2.2.1 in /workspace/wenet/venv/lib/python3.8/site-packages (from matplotlib) (2.4.7)\n",
      "Requirement already satisfied: six in /workspace/wenet/venv/lib/python3.8/site-packages (from cycler>=0.10->matplotlib) (1.15.0)\n",
      "Installing collected packages: kiwisolver, cycler, matplotlib\n",
      "Successfully installed cycler-0.10.0 kiwisolver-1.3.1 matplotlib-3.4.1\n"
     ]
    }
   ],
   "source": [
    "!pip install matplotlib\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "0f36d04f",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[<matplotlib.lines.Line2D at 0x7f0c39aa82e0>]"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYoAAAD4CAYAAADy46FuAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjQuMSwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/Z1A+gAAAACXBIWXMAAAsTAAALEwEAmpwYAAAqc0lEQVR4nO3deXxV1b338c8vCUkYkkAghJAEAhLQIJMEHHFCBa2KVkG0T7Wt1qet9ra1w9Xn3ufe1ld7b21tvVq1alut+mhJQK3Yqjig1SpCDgIyBiLTSZhCAglTyLSeP86GxjTDQZKc6ft+vXh5zjrrrLM2O+bL3mvv3zHnHCIiIu2JC/UEREQkvCkoRESkQwoKERHpkIJCREQ6pKAQEZEOJYR6Al1h0KBBLi8vL9TTEBGJKMuXL9/rnMvorF9UBEVeXh4+ny/U0xARiShmti2Yfjr1JCIiHVJQiIhIhxQUIiLSIQWFiIh0SEEhIiIdCioozGymmZWaWZmZ3d3G60lmVuS9vtTM8lq8do/XXmpmM1q0P2lme8xsTaux0s3sTTPb5P13wElsn4iInKROg8LM4oFHgMuBAuBGMyto1e1WYJ9zbhTwAHCf994CYC4wFpgJPOqNB/BHr621u4G3nXP5wNvecxERCZFgjiimAmXOuc3OuXpgHjCrVZ9ZwNPe4wXAdDMzr32ec+6oc24LUOaNh3PuPaC6jc9rOdbTwDXBb450p82VB3m3dE+opyEiPSyYoMgG/C2el3ttbfZxzjUCNcDAIN/bWqZzbqf3eBeQ2VYnM7vdzHxm5qusrAxiM+Rk3fS7pXzlqRLeWrc71FMRkR4U1ovZLvCtSm1+s5Jz7gnnXKFzrjAjo9M70OUkle05wK7aOgC+V7SSTysPhnhGItJTggmKCiC3xfMcr63NPmaWAKQBVUG+t7XdZpbljZUF6FxHGCj2lZMQZ7xy53kkJsRx+zM+DtQ1hHpaItIDggmKEiDfzEaYWSKBxemFrfosBG7xHl8PLPaOBhYCc72rokYA+cCyTj6v5Vi3AC8HMUfpRg1Nzbz4cTnTTxvMuJw0Hr7pDLZWHeZ7RatobtZX6YpEu06DwltzuBNYBKwHip1za83sXjO72uv2B2CgmZUBd+FdqeScWwsUA+uA14E7nHNNAGb2J2AJMMbMys3sVm+snwOXmtkm4BLvuYTQ4g172HuwnjmFgYPDs08ZyL9/4TTeWr+bB9/eFOLZiUh3s8A//CNbYWGhU/XY7nPb0yV8Ul7Dh3dfTEJ84N8Wzjl+uOATFiwv58G5E5k1sbNrFEQk3JjZcudcYWf9wnoxW0JvT20d75RWct3knOMhAWBm/Oza0zlzRDo/nP8JJVvbutJZRKKBgkI6tODjcpqa3fHTTi0lJcTz+Jcnk5Pem68/42PL3kMhmKGIdDcFhbTLOcd8XzlT89IZMahvm33690nkqa9MIc6Mrz61jOpD9T08SxHpbgoKaVfJ1n1s2XuIOVP++WiipeED+/K7myezo6aOrz/j40h9Uw/NUER6goJC2lXs89MvKYErxg3ptO/k4ek8eMNEVmzfxzefW059Y3MPzFBEeoKCQtp0oK6Bv36yk6smZNEnMbivVr98XBY/u3Yc75ZW8oP5usdCJFoE9xtAYs5fP9nJkYamNhexO3Lj1GHsP9zAfa9vIK13L+6dNZZAfUgRiVQKCmlTkc9P/uB+TMztf8Lv/eaFp7D/cD2Pv7eZ/n168f3LxnT9BEWkxygo5J9s2n2AFdv38+9fOO1zHw3cffmp1Bxp4DeLy0iMj+Pb0/O7eJYi0lMUFPJPin1+EuKMayZ9/rutAzfkjaO+sZlfvbkRM7jzYoWFSCRSUMhnBAoAVnDJaZkM6pd0UmPFxxm/nD0BgPvf2AgoLEQikYJCPuPt9XuoOlTPnCk5XTJe67AwM+64aFSXjC0iPUNBIZ8x3+cnMzWJ8/O77sugjoWFA365qJT6xma+e0m+roYSiRAKCjlud20d75Tu4RsXnPKZAoBdIT7OuH/2BBLijAff3kTNkQb+48oC4uIUFiLhTkEhxy1YXk6z44TvnQhWfJxx33XjSUnuxZMfbOFAXSP3XTeuy0NJRLqWgkKAYwUA/UwdkU5eOwUAu0JcnPF/rzyNtN69eOCtjRyoa+A3N00iKSG+2z5TRE6O/iknACzbUs3WqsPc0E1HEy2ZGd+5JJ//vKqAN9bt5qtPlVCr798WCVsKCgGg2FfuFQDM6rHP/Oq5I/j1nAks21LN9b/9kIr9R3rss0UkeAoK4UBdA6+u3slVE4bSO7FnTwF98Ywcnv7aVHbur+PaRz5gTUVNj36+iHROQSH8xSsAeEMn3zvRXc4dNYgF3zyHXvFxzHl8Ce9s2BOSeYhI2xQUQlGJn9GZ/ZiQkxayOYwZksJL3zqHkRl9ufXpEp5ZshXnVKZcJBwoKGLcxt0HWOnfz5zC3JDfADc4NZmi28/mojGD+Y+X13LPi6s52qhvyxMJNQVFjCsu8dMr3rj2JAoAdqW+SQk8cXMhd1x0CvNK/Nz0u6Xsqa0L9bREYpqCIobVNzbz0opAAcCBJ1kAsCvFxxk/nHEqj9x0But21HLVw39npX9/qKclErMUFDFs8YbdgQKAPXDvxOfxhfFZvPDNc0iICyxyF5f4Qz0lkZikoIhhxb5yhqQmc/7orisA2NUKhqbyyrfPo3D4AH70wid8v3gVh+sbQz0tkZiioIhRu2rqeLd0D9dNziY+zAvzpfdN5Nlbz+Rfpufz4opyZj38AWV7DoR6WiIxQ0ERo174OFAAcPbk8Dzt1Fp8nHHXpaN55mtTqT5Uz1W/+YCXVpSHeloiMUFBEYOccxT7/JzZzQUAu8O0/Axe/c40xuWk8b2iVfxowSoOHdWpKJHupKCIQUu3VLOt6nDI7sQ+WZmpyTx/25ncedEo5i8v54qH3mf5tn2hnpZI1FJQxKBin5+UpAQuP73nCgB2tYT4OH4wYwxFt59NY5Nj9mMf8us3N9LQ1BzqqYlEnaCCwsxmmlmpmZWZ2d1tvJ5kZkXe60vNLK/Fa/d47aVmNqOzMc1supl9bGYrzezvZqYvWO5CtccKAE7s+QKA3WHqiHRe++40rpmUzUNvb2L2Y0vYsvdQqKclElU6DQoziwceAS4HCoAbzaygVbdbgX3OuVHAA8B93nsLgLnAWGAm8KiZxXcy5m+BLznnJgLPA/9+Ulson/GXVTupa2juke+d6Cmpyb349ZyJPHzTJLbsPcQVD77PUx9soblZtaJEukIwRxRTgTLn3GbnXD0wD5jVqs8s4Gnv8QJgugUKB80C5jnnjjrntgBl3ngdjemAVO9xGrDj822atKXI52dMZgrjQ1gAsLtcOX4or393GmeOTOcnr6xjzuNL+LTyYKinJRLxggmKbKDlLbHlXlubfZxzjUANMLCD93Y05m3Aq2ZWDnwZ+HlbkzKz283MZ2a+ysrKIDZDSncdYJV/P3OmhL4AYHfJSuvNU1+Zwq9mT2DTnoNc/uD7/PbdT2nU2oXI5xaOi9nfA65wzuUATwG/bquTc+4J51yhc64wIyN87ywOJ8W+8CoA2F3MjOsm5/DmXedz0ZgM7nt9A9c++iHrd9aGemoiESmYoKgAWp7QzvHa2uxjZgkEThlVdfDeNtvNLAOY4Jxb6rUXAecEtSXSoWMFAC8tyCS9b2Kop9MjBqck89j/mswjN53Bjv1HuPI3f+e/Xl2v+y5ETlAwQVEC5JvZCDNLJLA4vbBVn4XALd7j64HFLvCtMwuBud5VUSOAfGBZB2PuA9LMbLQ31qXA+s+/eXLM2+t3U32ontlRtIgdDDPjC+OzeOuuC5hTmMMT721m+q/+xmurd+qLkUSClNBZB+dco5ndCSwC4oEnnXNrzexewOecWwj8AXjWzMqAagK/+PH6FQPrgEbgDudcE0BbY3rtXwdeMLNmAsHxtS7d4hhV5PMHCgDmx+ZpugF9E/nvL45ndmEu//bSGr753MdcMDqDn1w9NuLuThfpaRYN/6oqLCx0Pp8v1NMIWztrjnDuzxfzrQtH8YMZY0I9nZBrbGrm2Y+28as3NlLf1Mw3LjiFb1wwkj6Jnf67SSSqmNly51xhZ/3CcTFbutgLy70CgIU5oZ5KWEiIj+Or547g7e9fwMyxQ3jo7U1cfP/fePHjct17IdIGBUWUa252FPvKOWtkOsMH6hRLS5mpyTx04yQWfONsMlOTuKt4Fdc8+gG+rdWhnppIWFFQRLmlW6rZXh25BQB7QmFeOi9961weuGECe2qPcv1jS7jj+Y/xVx8O9dREwoJOyka5+T4/KcmRXQCwJ8TFGddOymHG2CE8/rfNPP7ep7y5djdfOmsYd1w0ikFh9J3iIj1NRxRRrLaugVfX7OTqCUNJ7hX5BQB7Qp/EBL536Wje+cGFXDspm6c/3MoFv3iHX7+5kQN1DaGenkhIKCii2CurdgQKAOq00wnLSuvNfdeP543vXcAFYzJ46O1NnP+Ld/j9+5upa2gK9fREepSCIooVl/g5dUgK47KjrwBgTxk1uB+PfmkyC+88l9Oz0/jpX9dz0f3v8vzS7dQ3qn6UxAYFRZTasKuWVeU1zCmM3gKAPWl8Tn+evfVMnr/tTDJTk/k/L63mwl++w7NLtnK0UUcYEt0UFFGquKScXvHGNVFeALCnnTNqEC996xye+dpUsvr35v++vJYLfvEuf/xgi05JSdRSUEShQAHAci4rGBIzBQB7kplx/ugMFnzjbJ677UyGpffhx6+sY5q3hnG4XkUHJbro8tgo9Nb63ew73KA7sbuZmXHuqEGcO2oQH22u4qG3N/HTv67n4XfKuPms4dx8Tp4uq5WooKCIQkUlfrLSkpkWowUAQ+GskQM5a+RAlm+r5rG/beahxWU8/t5mrp+cw9enjVThQYloCooos2P/Ed7bVMmdF40iPk6L2D1t8vB0fndzOmV7DvL79zcz31fO88u2M3PsEG4/fySThg0I9RRFTpiCIsq8sLwc52D2ZN07EUqjBvfj59eN565LR/PHD7fy/z7axmtrdjE1L51bzsnjsrGZ9IrXEqFEBpUZjyLNzY4L73+XnAG9ef7rZ4V6OtLCwaONzFu2naeXbMVffYQhqcl8+ezhzJ2Sy0CtY0iIqMx4DPpoSxXbqw8zJ8a+xS4S9EtK4LZpI3n3Bxfxu5sLGTW4H79cVMrZP1/M94tXsbq8JtRTFGmXTj1Fkfm+clKSE5h5+pBQT0XaER9nXFqQyaUFmZTtOcDTH27jhY/LeeHjcs4Y1p8vnz2cy0/PUm0uCSs69RQlao40MPVnbzG7MIefXjMu1NORE1Bb18ACXznPLNnK1qrDpPXuxbWTsrlx6jDGDEkJ9fQkigV76klHFFHilVU7ONrYzA2Fw0I9FTlBqcm9+Np5I/jKOXl8tLmK55dt57ml2/jjh1s5Y1h/bpw6jCvHD6V3oo4yJDR0RBElrn7479Q3NvPad6aptlMUqDp4lBc/ruBPJdvZXHmIlOQErpmYzQ1Tchk7NFX7WLqEjihiyPqdtXxSXsN/XlWgXyBRYmC/JL5+/khumzaCZVuq+dOy7RT5/Dz70TbGZKZw3eRsZk3MJjM1OdRTlRigoIgCxT4/ifFxXDNRBQCjjZlx5siBnDlyID8+XM8rn+zkxY/L+a9XN/Dz1zZwXn4G152RzWUFQ3RqSrqNgiLCHW1s4s8rKrh0bCYDVAAwqvXvk8iXzxrOl88azqeVB3np4wpeWlHBd+atpF9SAl8Yl8UXz8hmSl46cborX7qQgiLCvbVuD/sON+jeiRhzSkY/fjBjDHddOpqPtlTx4scV/OWTHRT5AnW+rhyfxZXjhzI+J02nI+WkaTE7wt385DLKdh/g/X+9WLWdYtzh+kbeWLubv3yyg79trKShyTEsvQ9XTcjiqglDGZOZotCQz9BidgzYsf8I72+q5NsqAChAn8QErpmUzTWTsqk53MCitbt45ZMdPPa3zTzyzqeMGtyPq8YP5coJWZyS0S/U05UIoqCIYAuOFQDUaSdpJa1PL+ZMyWXOlFz2HjzKa2t28ZdVO/iftzfywFsbOXVICpeNHcKMsZkUZOlyW+mYTj1FqOZmxwX3v8Ow9D48d5sKAEpwdtXU8erqnby+dhe+rdU0O8hN782MgiHMOH0IZwwboKPTGKJTT1Huo81V+KuP8IPLxoR6KhJBhqQl87XzRvC180aw9+BR3lq3m0Vrd/HMkm38/u9bGNQviUsLMpl5+hDOHjmQxATVDRUFRcQq9vlJTU5gxlgVAJTPZ1C/JOZOHcbcqcM4UNfAO6WVLFq7i5dXVvCnZdtJSUrg/NEZXHTqYC4ck6GvdY1hQQWFmc0EHgTigd87537e6vUk4BlgMlAF3OCc2+q9dg9wK9AE/ItzblFHY1rgZOlPgdnee37rnHvo5DYzutQcaeC1NbuYU5irKqPSJVKSe3H1hKFcPWEodQ1NfFC2lzfW7uad0j38dfVOzGBCTn+mnzqYi04drDIiMabToDCzeOAR4FKgHCgxs4XOuXUtut0K7HPOjTKzucB9wA1mVgDMBcYCQ4G3zGy09572xvwKkAuc6pxrNrPBXbGh0WThsQKAU7SILV0vuVc800/LZPppmTQ3O9btrOXt9XtYXLqHX725kV+9uZEhqclcdGoGF5+aybmjBtInUScnolkwe3cqUOac2wxgZvOAWUDLoJgF/Nh7vAB42DsymAXMc84dBbaYWZk3Hh2M+U3gJudcM4Bzbs/n37zoVFzi57SsVMYOTQ31VCTKxcUZp2encXp2Gt+5JJ89B+p4t7SSdzbsYeHKHfxpmZ/EhDim5qUzLX8Q54/O4NQhul8j2gQTFNmAv8XzcuDM9vo45xrNrAYY6LV/1Oq9xwoStTfmKQSORq4FKgmcrtrUelJmdjtwO8CwYbFTWnvdjlpWV9TwYxUAlBAYnJLMnMJc5hTmUt/YTMnWahZv2MP7myr579c28N+vbSAjJYlpowYxbfQgzhuVQUaK1jYiXTgeLyYBdc65QjP7IvAkMK11J+fcE8ATELg8tmenGDrHCgDOUgFACbHEhDjOHTWIc0cNAgKX3r6/qZL3N+3l3Y2VvLiiAoDTslI5P38Q0/IzKMwboHW1CBRMUFQQWDM4Jsdra6tPuZklAGkEFrU7em977eXAi97jl4CngphjTDja2MSfV1ZwmQoAShgakpbM7MJcZhfmHl/beG9TJe9v3MuTH2zh8fc2k5QQR2HeAM4eOZCzRg5kfE5/XYIbAYIJihIg38xGEPhlPhe4qVWfhcAtwBLgemCxc86Z2ULgeTP7NYHF7HxgGWAdjPln4CJgC3ABsPFzb12UeXPdbvarAKBEgJZrG9+6cBSHjjaybEs172/ay5LNVdz/RuB/69694gPBccpAzh45kHHZaSTEKzjCTadB4a053AksInAp65POubVmdi/gc84tBP4APOstVlcT+MWP16+YwCJ1I3CHc64JoK0xvY/8OfCcmX0POAjc1nWbG9mKSvxk9+99/FBfJFL0TUrgIu/SWoB9h+pZuqWKJZ9WsWRzFb94vRSAfkkJTDkeHIMoGJqqO8XDgEp4RIiK/Uc4777FfPvifO66dHTnbxCJIHsPHuWjzf8Ijs2VhwBISUrgjOEDmJI3gMK8dCbm9tcaRxdSCY8os8BXDsDsyTkhnolI1xvUL4krxw/lyvFDAdhdW8dHm6tYtqUa39Z9x09V9YoPnNKakpdO4fBAeKRrva7bKSgiQHOzY/5yP+eeMojc9D6hno5It8tMTWbWxOzjV/ftP1zP8m37KNm6D9/Wav74wVaeeG8zAKdk9A0Ehxcewwf20aXjXUxBEQGWbK6ifN8RfjhDBQAlNvXvk3j8bnGAuoYmVlfUULI1cMTx6uqdzCsJ3JqV3jeRibn9mZTbn0nDBjA+N43U5F6hnH7EU1BEABUAFPms5F7xTMlLZ0peOhA46t605yC+bdWs3L6fFf79LN4QKOpgBqMy+gXCY9gAJub2Z3RmP11ddQIUFGGu5nCgAODcKSoAKNKeuDhjzJAUxgxJ4UtnDgcCxTM/Kd/Piu37Wenfz1vrdzN/eWCtr09iPONz0piYGwiOCblpDElN1imrdigowtzCVRXUNzbr3gmRE5TWuxfT8jOYlp8BgHOObVWHWenfz4rt+1jp38/v399MY3Pgys9B/RI5PTuN8d79H+NyFB7HKCjCXJHPT0FWKqdnp4V6KiIRzczIG9SXvEF9uWZSYJG8rqGJtTtqWVNRw+qKGlaX1/Dexkq87GBQvyTGZacyLqc/47LTGJ+TRmZqcgi3IjQUFGFs7Y4a1lTU8pOrx4Z6KiJRKblXPJOHD2Dy8AHH247UN7FuZyA0VlfUsrpiP39rER4ZKUmM8446CrwqzjkDekf1kYeCIozN95WTmBDHrIlDQz0VkZjROzGeycPTmTw8/Xjb4fpG1u+s5ZPywJHHmooa3i3dczw8UpISOC0rldOyUjgtK5WCoamMzkyJmnVFBUWYqmto4qUVFcwYO4T+fXRDkUgo9UlM+KfwOFLfROnuA6zbUcv6nbWs21nLguXlHKpvAiDO4JSMfseD41iQDE6JvFNXCoow9ea63dQcaWBOoe7EFglHvRPjmZjbn4m5/Y+3NTc7tlcfZv3Of4TH8m37WLhqx/E+g/olcVpWCmMyUxg9JIXRmSnkD+5H36Tw/XUcvjOLccU+rwDgKSoAKBIp4uL+sWB++bis4+37D9ezfueB4+Gxfmctz360jaONzcf75Kb3ZkxmCvmZXohkpnDK4L4kJYT+9JWCIgyV7zvM38v28p3p+cSpcqZIxOvfJzFQEfeUgcfbmpod/urDlO4+wMZdByjdfYBNuw/ybmnl8Ut24+OMvIF9GO0Fx5ghKYzO7EfewL49esOggiIMLfBuCrpeBQBFolZ8i6OPllUX6hub2Vp1iI0tAmTDrgMsWrvr+OJ5YnwcIwb1ZdTgftx9+andXgNOQRFmmpsd833lnDdqEDkDVABQJNYkJsQdP4Jg/D/a6xqaKNtzMBAguw9Stucga3fU9Mg3BCoowsyHn1ZRsf8I/3r5qaGeioiEkeRe8ce/NbCnqSpWmCn2+Unr3YvLCjJDPRUREUBBEVZqDjfw+tpdXDNxaNTcqCMikU9BEUZePlYAcIoKAIpI+FBQhJGiEj9jh6YydqgKAIpI+FBQhIk1FTWs3VHLDTqaEJEwo6AIE/N9/kABwAnZoZ6KiMhnKCjCQF1DE39euYOZY4eQ1kff7Ssi4UVBEQbeOF4AUKedRCT8KCjCQHGJn5wBvTmnRR0YEZFwoaAIMX/1YT74dC+zJ+eqAKCIhCUFRYgdLwCo750QkTCloAih5mbHguWBAoDZ/XuHejoiIm1SUITQB5/upWL/ES1ii0hYU1CEULGvnP59enHZWBUAFJHwpaAIkf2H61m0dhfXTMwOi686FBFpT1BBYWYzzazUzMrM7O42Xk8ysyLv9aVmltfitXu89lIzm3ECYz5kZgc/53aFvZdX7ggUANRpJxEJc50GhZnFA48AlwMFwI1mVtCq263APufcKOAB4D7vvQXAXGAsMBN41MziOxvTzAqBASe5bWGtqMTP6dmpFAxNDfVUREQ6FMwRxVSgzDm32TlXD8wDZrXqMwt42nu8AJhuZua1z3POHXXObQHKvPHaHdMLkV8CPzq5TQtfaypqWLezlht0NCEiESCYoMgG/C2el3ttbfZxzjUCNcDADt7b0Zh3Agudczs7mpSZ3W5mPjPzVVZWBrEZ4aPYKwB4tQoAikgECKvFbDMbCswGftNZX+fcE865QudcYUZGRvdProvUNTTx5xUVXH66CgCKSGQIJigqgJbnSHK8tjb7mFkCkAZUdfDe9tonAaOAMjPbCvQxs7IgtyUiLFq7i9q6Ri1ii0jECCYoSoB8MxthZokEFqcXtuqzELjFe3w9sNg557z2ud5VUSOAfGBZe2M65/7qnBvinMtzzuUBh70F8qhR7POTm96bs0eqAKCIRIaEzjo45xrN7E5gERAPPOmcW2tm9wI+59xC4A/As96//qsJ/OLH61cMrAMagTucc00AbY3Z9ZsXXvzVh/mgrIq7Lh2tAoAiEjE6DQoA59yrwKut2v6jxeM6AmsLbb33Z8DPghmzjT79gplfpJi/vBwzuG6yCgCKSOQIq8XsaNbU7Fjg8zMtP0MFAEUkoigoesgHZXvZUVPHHJUTF5EIo6DoIcU+P/379OLSAhUAFJHIoqDoAfsO1fPG2t0qACgiEUlB0QNeXllBfZMKAIpIZFJQdDPnHEW+csZlp6kAoIhEJAVFN1tTUcv6nbXMmaKjCRGJTAqKblbs85OUEMfVE4aGeioiIp+LgqIb1TU08eeVXgHA3ioAKCKRSUHRjRat3cWBukaddhKRiKag6EZFJYECgGeNUAFAEYlcCopu4q8+zIefVjFncq4KAIpIRFNQdJP5Pr8KAIpIVFBQdIOmZseC5eWcn5/BUBUAFJEIp6DoBn8/XgBQi9giEvkUFN2g2OdnQJ9eXFIwONRTERE5aQqKLrbvUD1vrt3NNZNUAFBEooOCoou9tCJQAPAG3TshIlFCQdGFnHMU+/yMz0nj1CEqACgi0UFB0YVWV9SwYdcBLWKLSFRRUHShYwUAr1IBQBGJIgqKLlLX0MTLK3dwxbgsFQAUkaiioOgir6/xCgDqtJOIRBkFRRcpKvEzLL0PZ45ID/VURES6lIKiC2yvOsySzVXMKcxRAUARiToKii4wf7mfOBUAFJEopaA4SccLAI7OICtNBQBFJPooKE7S+5sq2akCgCISxRQUJ2m+r5z0volcclpmqKciItItFBQnofpQPW+s28U1E7NJTNBfpYhEp6B+u5nZTDMrNbMyM7u7jdeTzKzIe32pmeW1eO0er73UzGZ0NqaZPee1rzGzJ80sbO9ee2lFBQ1NTgUARSSqdRoUZhYPPAJcDhQAN5pZQatutwL7nHOjgAeA+7z3FgBzgbHATOBRM4vvZMzngFOBcUBv4LaT2sJu4pxjvs/PhJw0xgxJCfV0RES6TTBHFFOBMufcZudcPTAPmNWqzyzgae/xAmC6mZnXPs85d9Q5twUo88Zrd0zn3KvOAywDwvKa00/KvQKAOpoQkSgXTFBkA/4Wz8u9tjb7OOcagRpgYAfv7XRM75TTl4HX25qUmd1uZj4z81VWVgaxGV2r2OcnuZcKAIpI9AvnFdhHgfecc++39aJz7gnnXKFzrjAjI6NHJ3akvomFK3dwxelZpCaH7RKKiEiXSAiiTwXQ8vxKjtfWVp9yM0sA0oCqTt7b7phm9p9ABvC/g5hfj3t97U4OHG3UaScRiQnBHFGUAPlmNsLMEgksTi9s1WchcIv3+HpgsbfGsBCY610VNQLIJ7Du0O6YZnYbMAO40TnXfHKb1z2KSvwMH6gCgCISGzo9onDONZrZncAiIB540jm31szuBXzOuYXAH4BnzawMqCbwix+vXzGwDmgE7nDONQG0Nab3kY8B24AlgfVwXnTO3dtlW3yStlUd4qPN1fxwxhi8+YmIRLVgTj3hnHsVeLVV23+0eFwHzG7nvT8DfhbMmF57UHMKlfm+8kABwDPC8mIsEZEuF86L2WHnWAHAC0ZnMCQtOdTTERHpEQqKE/Depkp21aoAoIjEFgXFCZjv85PeN5HpKgAoIjFEQRGkqoNHeXPdbq6dpAKAIhJb9BsvSMcKAOq0k4jEGgVFEJxzFPv8TMjtrwKAIhJzFBRBWFVew8bdB7lBRxMiEoMUFEH4RwHArFBPRUSkxykoOnGkvolXVu7ginFZpKgAoIjEIAVFJ15bEygAqNNOIhKrFBSdKCrxkzewD1NVAFBEYpSCogNb9x5i6ZZqZhfmqgCgiMQsBUUH5i/3qwCgiMQ8BUU7jhUAvHDMYBUAFJGYpqBox3sbK9lde5Q5hTqaEJHYpqBoR1GJn4F9E7n4VBUAFJHYpqBoQ9XBo7y1XgUARURAQdGml1ZU0NjsmDNF906IiCgoWnHOUVTiZ2Juf0ZnqgCgiIiCopWV/v1s2nOQG3Q0ISICKCj+SbGvnN694rlyvAoAioiAguIzDtc38soqFQAUEWlJQdHCa6t3cfBoo047iYi0oKBoocjnZ8SgvkzJGxDqqYiIhA0FhWfL3kMs21LN7MIcFQAUEWlBQeGZ71MBQBGRtigogMamZl74uJyLxgwmM1UFAEVEWlJQAO9tChQAnK1vsRMR+ScKCgIFAAf1S2T6aYNDPRURkbAT80Gx9+BR3l6/h2snZdMrPub/OkRE/knM/2Z86eNAAUDdOyEi0raggsLMZppZqZmVmdndbbyeZGZF3utLzSyvxWv3eO2lZjajszHNbIQ3Rpk3ZuJJbmO7nHMU+/ycMaw/owarAKCISFs6DQoziwceAS4HCoAbzaygVbdbgX3OuVHAA8B93nsLgLnAWGAm8KiZxXcy5n3AA95Y+7yxu8UKrwDgHC1ii4i0K5gjiqlAmXNus3OuHpgHzGrVZxbwtPd4ATDdAnetzQLmOeeOOue2AGXeeG2O6b3nYm8MvDGv+dxb14n5Pn+gAOCEod31ESIiES+YoMgG/C2el3ttbfZxzjUCNcDADt7bXvtAYL83RnufBYCZ3W5mPjPzVVZWBrEZ/2xYel++cm4e/ZISPtf7RURiQcT+hnTOPQE8AVBYWOg+zxjfvPCULp2TiEg0CuaIogJoeRI/x2trs4+ZJQBpQFUH722vvQro743R3meJiEgPCiYoSoB872qkRAKL0wtb9VkI3OI9vh5Y7JxzXvtc76qoEUA+sKy9Mb33vOONgTfmy59/80RE5GR1eurJOddoZncCi4B44Enn3FozuxfwOecWAn8AnjWzMqCawC9+vH7FwDqgEbjDOdcE0NaY3kf+KzDPzH4KrPDGFhGRELHAP+IjW2FhofP5fKGehohIRDGz5c65ws76xfyd2SIi0jEFhYiIdEhBISIiHVJQiIhIh6JiMdvMKoFtn/Ptg4C9XTidSKBtjg3a5uh3sts73DmX0VmnqAiKk2FmvmBW/aOJtjk2aJujX09tr049iYhIhxQUIiLSIQWFV1gwxmibY4O2Ofr1yPbG/BqFiIh0TEcUIiLSIQWFiIh0KKaDwsxmmlmpmZWZ2d2hns+JMLNcM3vHzNaZ2Voz+47Xnm5mb5rZJu+/A7x2M7OHvG39xMzOaDHWLV7/TWZ2S4v2yWa22nvPQ95X1Yac973rK8zsL97zEWa21JtnkVe6Hq+8fZHXvtTM8lqMcY/XXmpmM1q0h93PhJn1N7MFZrbBzNab2dnRvp/N7Hvez/UaM/uTmSVH2342syfNbI+ZrWnR1u37tb3P6JBzLib/EChv/ikwEkgEVgEFoZ7XCcw/CzjDe5wCbAQKgF8Ad3vtdwP3eY+vAF4DDDgLWOq1pwObvf8O8B4P8F5b5vU1772Xh3q7vXndBTwP/MV7XgzM9R4/BnzTe/wt4DHv8VygyHtc4O3vJGCE93MQH64/EwS+O/4273Ei0D+a9zOBrz/eAvRusX+/Em37GTgfOANY06Kt2/dre5/R4VxD/T9BCH8YzwYWtXh+D3BPqOd1EtvzMnApUApkeW1ZQKn3+HHgxhb9S73XbwQeb9H+uNeWBWxo0f6ZfiHczhzgbeBi4C/e/wR7gYTW+5XA952c7T1O8PpZ6319rF84/kwQ+LbILXgXnrTef9G4nwkEhd/75Zfg7ecZ0bifgTw+GxTdvl/b+4yO/sTyqadjP4zHlHttEcc71J4ELAUynXM7vZd2AZne4/a2t6P28jbaQ+1/gB8Bzd7zgcB+51yj97zlPI9vm/d6jdf/RP8uQmkEUAk85Z1u+72Z9SWK97NzrgK4H9gO7CSw35YT3fv5mJ7Yr+19RrtiOSiigpn1A14Avuucq235mgv8kyFqrn82syuBPc655aGeSw9KIHB64rfOuUnAIQKnC46Lwv08AJhFICSHAn2BmSGdVAj0xH4N9jNiOSgqgNwWz3O8tohhZr0IhMRzzrkXvebdZpblvZ4F7PHa29vejtpz2mgPpXOBq81sKzCPwOmnB4H+Znbsa31bzvP4tnmvpwFVnPjfRSiVA+XOuaXe8wUEgiOa9/MlwBbnXKVzrgF4kcC+j+b9fExP7Nf2PqNdsRwUJUC+dyVFIoFFsIUhnlPQvCsY/gCsd879usVLC4FjVz7cQmDt4lj7zd7VE2cBNd7h5yLgMjMb4P1L7jIC5293ArVmdpb3WTe3GCsknHP3OOdynHN5BPbXYufcl4B3gOu9bq23+djfxfVef+e1z/WulhkB5BNY+Au7nwnn3C7Ab2ZjvKbpBL6DPmr3M4FTTmeZWR9vTse2OWr3cws9sV/b+4z2hXLRKtR/CFxJsJHAFRD/Fur5nODczyNwyPgJsNL7cwWBc7NvA5uAt4B0r78Bj3jbuhoobDHW14Ay789XW7QXAmu89zxMqwXVEG//hfzjqqeRBH4BlAHzgSSvPdl7Xua9PrLF+//N265SWlzlE44/E8BEwOft6z8TuLolqvcz8BNggzevZwlcuRRV+xn4E4E1mAYCR4639sR+be8zOvqjEh4iItKhWD71JCIiQVBQiIhIhxQUIiLSIQWFiIh0SEEhIiIdUlCIiEiHFBQiItKh/w/uhegfvR+Q7QAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "xs = list(range(100000))\n",
    "plt.plot(xs, lrs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "4f4e282c",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/workspace/wenet/venv/lib/python3.8/site-packages/ipykernel/ipkernel.py:283: DeprecationWarning: `should_run_async` will not call `transform_cell` automatically in the future. Please pass the result to `transformed_cell` argument and any exception that happen during thetransform in `preprocessing_exc_tuple` in IPython 7.17 and above.\n",
      "  and should_run_async(code)\n"
     ]
    }
   ],
   "source": [
    "from typing import Union\n",
    "\n",
    "from paddle.optimizer.lr import LRScheduler\n",
    "from typeguard import check_argument_types\n",
    "\n",
    "class WarmupLR(LRScheduler):\n",
    "    \"\"\"The WarmupLR scheduler\n",
    "    This scheduler is almost same as NoamLR Scheduler except for following\n",
    "    difference:\n",
    "    NoamLR:\n",
    "        lr = optimizer.lr * model_size ** -0.5\n",
    "             * min(step ** -0.5, step * warmup_step ** -1.5)\n",
    "    WarmupLR:\n",
    "        lr = optimizer.lr * warmup_step ** 0.5\n",
    "             * min(step ** -0.5, step * warmup_step ** -1.5)\n",
    "    Note that the maximum lr equals to optimizer.lr in this scheduler.\n",
    "    \"\"\"\n",
    "\n",
    "    def __init__(self,\n",
    "                 warmup_steps: Union[int, float]=25000,\n",
    "                 learning_rate=1.0,\n",
    "                 last_epoch=-1,\n",
    "                 verbose=False):\n",
    "        assert check_argument_types()\n",
    "        self.warmup_steps = warmup_steps\n",
    "        super().__init__(learning_rate, last_epoch, verbose)\n",
    "\n",
    "    def __repr__(self):\n",
    "        return f\"{self.__class__.__name__}(warmup_steps={self.warmup_steps})\"\n",
    "\n",
    "    def get_lr(self):\n",
    "        step_num = self.last_epoch + 1\n",
    "        return self.base_lr * self.warmup_steps**0.5 * min(\n",
    "            step_num**-0.5, step_num * self.warmup_steps**-1.5)\n",
    "\n",
    "    def set_step(self, step: int):\n",
    "        self.step(step)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "8c40b202",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "-1\n"
     ]
    }
   ],
   "source": [
    "sc = WarmupLR(warmup_steps=25000, learning_rate=0.001)\n",
    "print(step)\n",
    "#sc.set_step(step)\n",
    "sc.set_step(0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "ecbc7e37",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[<matplotlib.lines.Line2D at 0x7f0ba6dd9c40>]"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "lrs=[]\n",
    "for i in range(100000):\n",
    "    sc.step()\n",
    "    lrs.append(sc.get_lr())\n",
    "xs = list(range(100000))\n",
    "plt.plot(xs, lrs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "e613fe16",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "f0fd9f40",
   "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": "3.8.0"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}