Transformer-for-nlp-task.ipynb

{
  "cells": [
    {
      "cell_type": "code",
      "execution_count": 14,
      "metadata": {
        "id": "i8NdEmymA3Nb"
      },
      "outputs": [],
      "source": [
        "%matplotlib inline"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "!pip install torchdata"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "4F38JOoMKO_Z",
        "outputId": "92c2727a-193a-4871-bb37-19389f5be402"
      },
      "execution_count": 17,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Looking in indexes: https://pypi.org/simple, https://us-python.pkg.dev/colab-wheels/public/simple/\n",
            "Requirement already satisfied: torchdata in /usr/local/lib/python3.7/dist-packages (0.3.0)\n",
            "Requirement already satisfied: urllib3>=1.25 in /usr/local/lib/python3.7/dist-packages (from torchdata) (1.25.11)\n",
            "Requirement already satisfied: torch==1.11.0 in /usr/local/lib/python3.7/dist-packages (from torchdata) (1.11.0+cu113)\n",
            "Requirement already satisfied: requests in /usr/local/lib/python3.7/dist-packages (from torchdata) (2.23.0)\n",
            "Requirement already satisfied: typing-extensions in /usr/local/lib/python3.7/dist-packages (from torch==1.11.0->torchdata) (4.2.0)\n",
            "Requirement already satisfied: certifi>=2017.4.17 in /usr/local/lib/python3.7/dist-packages (from requests->torchdata) (2022.5.18.1)\n",
            "Requirement already satisfied: chardet<4,>=3.0.2 in /usr/local/lib/python3.7/dist-packages (from requests->torchdata) (3.0.4)\n",
            "Requirement already satisfied: idna<3,>=2.5 in /usr/local/lib/python3.7/dist-packages (from requests->torchdata) (2.10)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 15,
      "metadata": {
        "id": "aeK0EKLKA3Ni"
      },
      "outputs": [],
      "source": [
        "import math\n",
        "from typing import Tuple\n",
        "\n",
        "import torch\n",
        "from torch import nn, Tensor\n",
        "import torch.nn.functional as F\n",
        "from torch.nn import TransformerEncoder, TransformerEncoderLayer\n",
        "from torch.utils.data import dataset\n",
        "\n",
        "class TransformerModel(nn.Module):\n",
        "\n",
        "    def __init__(self, ntoken: int, d_model: int, nhead: int, d_hid: int,\n",
        "                 nlayers: int, dropout: float = 0.5):\n",
        "        super().__init__()\n",
        "        self.model_type = 'Transformer'\n",
        "        self.pos_encoder = PositionalEncoding(d_model, dropout)\n",
        "        encoder_layers = TransformerEncoderLayer(d_model, nhead, d_hid, dropout)\n",
        "        self.transformer_encoder = TransformerEncoder(encoder_layers, nlayers)\n",
        "        self.encoder = nn.Embedding(ntoken, d_model)\n",
        "        self.d_model = d_model\n",
        "        self.decoder = nn.Linear(d_model, ntoken)\n",
        "\n",
        "        self.init_weights()\n",
        "\n",
        "    def init_weights(self) -> None:\n",
        "        initrange = 0.1\n",
        "        self.encoder.weight.data.uniform_(-initrange, initrange)\n",
        "        self.decoder.bias.data.zero_()\n",
        "        self.decoder.weight.data.uniform_(-initrange, initrange)\n",
        "\n",
        "    def forward(self, src: Tensor, src_mask: Tensor) -> Tensor:\n",
        "        src = self.encoder(src) * math.sqrt(self.d_model)\n",
        "        src = self.pos_encoder(src)\n",
        "        output = self.transformer_encoder(src, src_mask)\n",
        "        output = self.decoder(output)\n",
        "        return output\n",
        "\n",
        "\n",
        "def generate_square_subsequent_mask(sz: int) -> Tensor:\n",
        "    return torch.triu(torch.ones(sz, sz) * float('-inf'), diagonal=1)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 16,
      "metadata": {
        "id": "BFUD5f3GA3Nj"
      },
      "outputs": [],
      "source": [
        "class PositionalEncoding(nn.Module):\n",
        "\n",
        "    def __init__(self, d_model: int, dropout: float = 0.1, max_len: int = 5000):\n",
        "        super().__init__()\n",
        "        self.dropout = nn.Dropout(p=dropout)\n",
        "\n",
        "        position = torch.arange(max_len).unsqueeze(1)\n",
        "        div_term = torch.exp(torch.arange(0, d_model, 2) * (-math.log(10000.0) / d_model))\n",
        "        pe = torch.zeros(max_len, 1, d_model)\n",
        "        pe[:, 0, 0::2] = torch.sin(position * div_term)\n",
        "        pe[:, 0, 1::2] = torch.cos(position * div_term)\n",
        "        self.register_buffer('pe', pe)\n",
        "\n",
        "    def forward(self, x: Tensor) -> Tensor:\n",
        "        x = x + self.pe[:x.size(0)]\n",
        "        return self.dropout(x)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 18,
      "metadata": {
        "id": "rR3OdfrBA3Nk"
      },
      "outputs": [],
      "source": [
        "from torchtext.datasets import IMDB\n",
        "from torchtext.data.utils import get_tokenizer\n",
        "from torchtext.vocab import build_vocab_from_iterator\n",
        "from typing import Iterable, List\n",
        "\n",
        "train_iter = IMDB(split='train')\n",
        "\n",
        "\n",
        "tokenizer = get_tokenizer('basic_english')\n",
        "vocab = build_vocab_from_iterator(map(tokenizer, train_iter), specials=['<unk>'])\n",
        "vocab.set_default_index(vocab['<unk>']) \n",
        "\n",
        "def data_process(raw_text_iter: dataset.IterableDataset) -> Tensor:\n",
        "    \"\"\"Converts raw text into a flat Tensor.\"\"\"\n",
        "    data = [torch.tensor(vocab(tokenizer(item)), dtype=torch.long) for item in raw_text_iter]\n",
        "    return torch.cat(tuple(filter(lambda t: t.numel() > 0, data)))\n",
        "\n",
        "train_iter, val_iter, test_iter = IMDB()\n",
        "train_data = data_process(train_iter)\n",
        "val_data = data_process(val_iter)\n",
        "test_data = data_process(test_iter)\n",
        "\n",
        "device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')\n",
        "\n",
        "def batchify(data: Tensor, bsz: int) -> Tensor:\n",
        "    seq_len = data.size(0) // bsz\n",
        "    data = data[:seq_len * bsz]\n",
        "    data = data.view(bsz, seq_len).t().contiguous()\n",
        "    return data.to(device)\n",
        "\n",
        "batch_size = 20\n",
        "eval_batch_size = 10\n",
        "train_data = batchify(train_data, batch_size)\n",
        "val_data = batchify(val_data, eval_batch_size)\n",
        "test_data = batchify(test_data, eval_batch_size)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 19,
      "metadata": {
        "id": "0qXiRa--A3Nl"
      },
      "outputs": [],
      "source": [
        "bptt = 35\n",
        "def get_batch(source: Tensor, i: int) -> Tuple[Tensor, Tensor]:\n",
        "    seq_len = min(bptt, len(source) - 1 - i)\n",
        "    data = source[i:i+seq_len]\n",
        "    target = source[i+1:i+1+seq_len].reshape(-1)\n",
        "    return data, target"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "#@title Training params\n",
        "emsize = 200 #@param {type:\"number\"}\n",
        "d_hid = 200 #@param {type:\"number\"}\n",
        "dropout = 0.2  #@param {type:\"slider\", min:0, max:1, step:0.05}\n",
        "nhead = 2 #@param {type:\"slider\", min:1, max:8, step:1}\n",
        "nlayers = 2 #@param {type:\"slider\", min:1, max:8, step:1}\n",
        "lr = 0.0001 #@param {type:\"slider\", min:0.0001, max:5.0, step:0.0001}\n",
        "epochs = 3 #@param {type:\"slider\", min:1, max:1000, step:1}"
      ],
      "metadata": {
        "id": "rbMXpphH7O75"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "execution_count": 20,
      "metadata": {
        "id": "NekFiFMRA3Nm"
      },
      "outputs": [],
      "source": [
        "ntokens = len(vocab)\n",
        "criterion = nn.CrossEntropyLoss()\n",
        "model = TransformerModel(ntokens, emsize, nhead, d_hid, nlayers, dropout).to(device)\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 21,
      "metadata": {
        "id": "I3UaUX6rA3Nn"
      },
      "outputs": [],
      "source": [
        "import copy\n",
        "import time\n",
        "\n",
        "optimizer = torch.optim.SGD(model.parameters(), lr=lr)\n",
        "scheduler = torch.optim.lr_scheduler.StepLR(optimizer, 1.0, gamma=0.95)\n",
        "\n",
        "def train(model: nn.Module) -> None:\n",
        "    model.train()  # turn on train mode\n",
        "    total_loss = 0.\n",
        "    log_interval = 200\n",
        "    start_time = time.time()\n",
        "    src_mask = generate_square_subsequent_mask(bptt).to(device)\n",
        "\n",
        "    num_batches = len(train_data) // bptt\n",
        "    for batch, i in enumerate(range(0, train_data.size(0) - 1, bptt)):\n",
        "        data, targets = get_batch(train_data, i)\n",
        "        batch_size = data.size(0)\n",
        "        if batch_size != bptt:  # only on last batch\n",
        "            src_mask = src_mask[:batch_size, :batch_size]\n",
        "        output = model(data, src_mask)\n",
        "        loss = criterion(output.view(-1, ntokens), targets)\n",
        "\n",
        "        optimizer.zero_grad()\n",
        "        loss.backward()\n",
        "        torch.nn.utils.clip_grad_norm_(model.parameters(), 0.5)\n",
        "        optimizer.step()\n",
        "\n",
        "        total_loss += loss.item()\n",
        "        if batch % log_interval == 0 and batch > 0:\n",
        "            lr = scheduler.get_last_lr()[0]\n",
        "            ms_per_batch = (time.time() - start_time) * 1000 / log_interval\n",
        "            cur_loss = total_loss / log_interval\n",
        "            ppl = math.exp(cur_loss)\n",
        "            print(f'| epoch {epoch:3d} | {batch:5d}/{num_batches:5d} batches | '\n",
        "                  f'lr {lr:02.2f} | ms/batch {ms_per_batch:5.2f} | '\n",
        "                  f'loss {cur_loss:5.2f} | ppl {ppl:8.2f}')\n",
        "            total_loss = 0\n",
        "            start_time = time.time()\n",
        "\n",
        "def evaluate(model: nn.Module, eval_data: Tensor) -> float:\n",
        "    model.eval()  # turn on evaluation mode\n",
        "    total_loss = 0.\n",
        "    src_mask = generate_square_subsequent_mask(bptt).to(device)\n",
        "    with torch.no_grad():\n",
        "        for i in range(0, eval_data.size(0) - 1, bptt):\n",
        "            data, targets = get_batch(eval_data, i)\n",
        "            batch_size = data.size(0)\n",
        "            if batch_size != bptt:\n",
        "                src_mask = src_mask[:batch_size, :batch_size]\n",
        "            output = model(data, src_mask)\n",
        "            output_flat = output.view(-1, ntokens)\n",
        "            total_loss += batch_size * criterion(output_flat, targets).item()\n",
        "    return total_loss / (len(eval_data) - 1)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 22,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 972
        },
        "id": "qLcC1nKEA3Nn",
        "outputId": "158c41ea-c8e8-4ba6-b1e3-b67fedfb5443"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "| epoch   1 |   200/145063 batches | lr 5.00 | ms/batch 95.66 | loss  9.69 | ppl 16129.10\n",
            "| epoch   1 |   400/145063 batches | lr 5.00 | ms/batch 95.28 | loss  8.04 | ppl  3102.31\n",
            "| epoch   1 |   600/145063 batches | lr 5.00 | ms/batch 96.58 | loss  7.41 | ppl  1646.56\n",
            "| epoch   1 |   800/145063 batches | lr 5.00 | ms/batch 97.98 | loss  7.11 | ppl  1227.39\n",
            "| epoch   1 |  1000/145063 batches | lr 5.00 | ms/batch 99.72 | loss  6.93 | ppl  1017.96\n",
            "| epoch   1 |  1200/145063 batches | lr 5.00 | ms/batch 101.44 | loss  6.83 | ppl   922.97\n",
            "| epoch   1 |  1400/145063 batches | lr 5.00 | ms/batch 101.75 | loss  6.83 | ppl   928.31\n",
            "| epoch   1 |  1600/145063 batches | lr 5.00 | ms/batch 101.32 | loss  6.69 | ppl   806.08\n",
            "| epoch   1 |  1800/145063 batches | lr 5.00 | ms/batch 100.95 | loss  6.65 | ppl   773.62\n",
            "| epoch   1 |  2000/145063 batches | lr 5.00 | ms/batch 101.48 | loss  6.58 | ppl   717.20\n",
            "| epoch   1 |  2200/145063 batches | lr 5.00 | ms/batch 101.58 | loss  6.52 | ppl   680.18\n",
            "| epoch   1 |  2400/145063 batches | lr 5.00 | ms/batch 101.57 | loss  6.43 | ppl   622.69\n",
            "| epoch   1 |  2600/145063 batches | lr 5.00 | ms/batch 101.46 | loss  6.46 | ppl   642.07\n",
            "| epoch   1 |  2800/145063 batches | lr 5.00 | ms/batch 101.60 | loss  6.47 | ppl   643.27\n",
            "| epoch   1 |  3000/145063 batches | lr 5.00 | ms/batch 102.33 | loss  6.41 | ppl   607.36\n",
            "| epoch   1 |  3200/145063 batches | lr 5.00 | ms/batch 101.77 | loss  6.39 | ppl   593.54\n",
            "| epoch   1 |  3400/145063 batches | lr 5.00 | ms/batch 102.15 | loss  6.44 | ppl   626.15\n",
            "| epoch   1 |  3600/145063 batches | lr 5.00 | ms/batch 101.61 | loss  6.35 | ppl   571.14\n",
            "| epoch   1 |  3800/145063 batches | lr 5.00 | ms/batch 101.58 | loss  6.29 | ppl   540.66\n",
            "| epoch   1 |  4000/145063 batches | lr 5.00 | ms/batch 101.78 | loss  6.24 | ppl   513.11\n",
            "| epoch   1 |  4200/145063 batches | lr 5.00 | ms/batch 101.83 | loss  6.31 | ppl   549.64\n",
            "| epoch   1 |  4400/145063 batches | lr 5.00 | ms/batch 101.78 | loss  6.28 | ppl   534.39\n",
            "| epoch   1 |  4600/145063 batches | lr 5.00 | ms/batch 101.99 | loss  6.27 | ppl   527.86\n",
            "| epoch   1 |  4800/145063 batches | lr 5.00 | ms/batch 101.71 | loss  6.26 | ppl   524.82\n",
            "| epoch   1 |  5000/145063 batches | lr 5.00 | ms/batch 101.55 | loss  6.28 | ppl   533.22\n",
            "| epoch   1 |  5200/145063 batches | lr 5.00 | ms/batch 101.70 | loss  6.19 | ppl   486.69\n",
            "| epoch   1 |  5400/145063 batches | lr 5.00 | ms/batch 101.69 | loss  6.12 | ppl   454.83\n",
            "| epoch   1 |  5600/145063 batches | lr 5.00 | ms/batch 101.63 | loss  6.17 | ppl   478.07\n",
            "| epoch   1 |  5800/145063 batches | lr 5.00 | ms/batch 101.72 | loss  6.09 | ppl   442.78\n",
            "| epoch   1 |  6000/145063 batches | lr 5.00 | ms/batch 101.74 | loss  6.25 | ppl   517.07\n",
            "| epoch   1 |  6200/145063 batches | lr 5.00 | ms/batch 101.59 | loss  6.15 | ppl   467.97\n",
            "| epoch   1 |  6400/145063 batches | lr 5.00 | ms/batch 101.54 | loss  6.18 | ppl   483.69\n",
            "| epoch   1 |  6600/145063 batches | lr 5.00 | ms/batch 101.66 | loss  6.16 | ppl   474.11\n",
            "| epoch   1 |  6800/145063 batches | lr 5.00 | ms/batch 101.56 | loss  6.16 | ppl   472.62\n"
          ]
        },
        {
          "output_type": "error",
          "ename": "KeyboardInterrupt",
          "evalue": "ignored",
          "traceback": [
            "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
            "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
            "\u001b[0;32m<ipython-input-22-eea392489cd8>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m      5\u001b[0m \u001b[0;32mfor\u001b[0m \u001b[0mepoch\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mrange\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mepochs\u001b[0m \u001b[0;34m+\u001b[0m \u001b[0;36m1\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      6\u001b[0m     \u001b[0mepoch_start_time\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mtime\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mtime\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----> 7\u001b[0;31m     \u001b[0mtrain\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mmodel\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      8\u001b[0m     \u001b[0mval_loss\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mevaluate\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mmodel\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mval_data\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      9\u001b[0m     \u001b[0mval_ppl\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mmath\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mexp\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mval_loss\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m<ipython-input-21-7281ae317664>\u001b[0m in \u001b[0;36mtrain\u001b[0;34m(model)\u001b[0m\n\u001b[1;32m     28\u001b[0m         \u001b[0moptimizer\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mstep\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     29\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 30\u001b[0;31m         \u001b[0mtotal_loss\u001b[0m \u001b[0;34m+=\u001b[0m \u001b[0mloss\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mitem\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     31\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mbatch\u001b[0m \u001b[0;34m%\u001b[0m \u001b[0mlog_interval\u001b[0m \u001b[0;34m==\u001b[0m \u001b[0;36m0\u001b[0m \u001b[0;32mand\u001b[0m \u001b[0mbatch\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     32\u001b[0m             \u001b[0mlr\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mscheduler\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mget_last_lr\u001b[0m\u001b[0;34m(\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[0;31mKeyboardInterrupt\u001b[0m: "
          ]
        }
      ],
      "source": [
        "best_val_loss = float('inf')\n",
        "best_model = None\n",
        "\n",
        "for epoch in range(1, epochs + 1):\n",
        "    epoch_start_time = time.time()\n",
        "    train(model)\n",
        "    val_loss = evaluate(model, val_data)\n",
        "    val_ppl = math.exp(val_loss)\n",
        "    elapsed = time.time() - epoch_start_time\n",
        "    print('-' * 89)\n",
        "    print(f'| end of epoch {epoch:3d} | time: {elapsed:5.2f}s | '\n",
        "          f'valid loss {val_loss:5.2f} | valid ppl {val_ppl:8.2f}')\n",
        "    print('-' * 89)\n",
        "\n",
        "    if val_loss < best_val_loss:\n",
        "        best_val_loss = val_loss\n",
        "        best_model = copy.deepcopy(model)\n",
        "\n",
        "    scheduler.step()"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "Z0SWZNqCA3No"
      },
      "outputs": [],
      "source": [
        "test_loss = evaluate(best_model, test_data)\n",
        "test_ppl = math.exp(test_loss)\n",
        "print('=' * 89)\n",
        "print(f'| End of training | test loss {test_loss:5.2f} | '\n",
        "      f'test ppl {test_ppl:8.2f}')\n",
        "print('=' * 89)"
      ]
    }
  ],
  "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": "3.7.13"
    },
    "colab": {
      "name": "transformer_tutorial.ipynb",
      "provenance": []
    },
    "accelerator": "GPU"
  },
  "nbformat": 4,
  "nbformat_minor": 0
}