Transformer_Scratch.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "provenance": []
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    }
  },
  "cells": [
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "l1vwWO1khBH-"
      },
      "outputs": [],
      "source": [
        "import numpy as np\n",
        "def selfAttention(Q,K,V):\n",
        "  dk = np.dim(K)\n",
        "  Attention = (np.matmul(Q,K))/dk\n",
        "\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "arr = np.array([[1,2,3],[3,4,5],[6,7,8]])\n"
      ],
      "metadata": {
        "id": "7kczUvYZheHx"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "import numpy as np\n",
        "import matplotlib.pyplot as plt\n",
        "def evenpos(i,j,d_model):\n",
        "  return np.sin(i/10000**(2*j/d_model))\n",
        "def oddpos(i,j,d_model):\n",
        "  return np.cos(i/10000**(2*j/d_model))\n",
        "\n",
        "def PositionEncoding(max_len,d_model,n=10000):\n",
        "  arr = np.zeros([max_len,d_model])\n",
        "  for k in range(max_len):\n",
        "    for i in range(0,d_model,2):\n",
        "      denominator = np.power(n,2*i/d_model)\n",
        "      arr[k,i] = evenpos(k,i,d_model)\n",
        "      arr[k,i+1] = oddpos(k,i,d_model)\n",
        "  return arr\n",
        "\n"
      ],
      "metadata": {
        "id": "jjiypAHIiCKY"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "arr = PositionEncoding(max_len = 100, d_model = 512, n = 10000)\n",
        "print(arr)\n",
        "cax = plt.matshow(arr)\n",
        "plt.colorbar(cax)\n",
        "plt.show()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 516
        },
        "id": "W-pcKIk-pHUZ",
        "outputId": "bb259bab-017a-4e98-c7ba-081c32afd23d"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "[[ 0.00000000e+00  1.00000000e+00  0.00000000e+00 ...  1.00000000e+00\n",
            "   0.00000000e+00  1.00000000e+00]\n",
            " [ 8.41470985e-01  5.40302306e-01  8.01961795e-01 ...  1.00000000e+00\n",
            "   1.07460783e-08  1.00000000e+00]\n",
            " [ 9.09297427e-01 -4.16146837e-01  9.58144376e-01 ...  1.00000000e+00\n",
            "   2.14921566e-08  1.00000000e+00]\n",
            " ...\n",
            " [ 3.79607739e-01 -9.25147537e-01  7.45109484e-01 ...  1.00000000e+00\n",
            "   1.04236959e-06  1.00000000e+00]\n",
            " [-5.73381872e-01 -8.19288245e-01 -8.97521193e-02 ...  1.00000000e+00\n",
            "   1.05311567e-06  1.00000000e+00]\n",
            " [-9.99206834e-01  3.98208804e-02 -8.52340887e-01 ...  1.00000000e+00\n",
            "   1.06386175e-06  1.00000000e+00]]\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 1600x312.5 with 2 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "import math as mt\n",
        "import torch.nn as nn\n",
        "from torch import Tensor\n",
        "\n",
        "class Embedding(nn.Module):\n",
        "  def __init__(self,vocab_size: int, dim_embed: int)-> None:\n",
        "    super().__init__()\n",
        "    self.embedding = nn.Embedding(vocab_size,dim_embed)\n",
        "    self.sqrt_dim_embed = mt.sqrt(dim_embed)\n",
        "  def forward(self, x: Tensor) -> Tensor:\n",
        "    x = self.embedding(x.long())\n",
        "    x = self.sqrt_dim_embed(x)\n",
        "    return x\n",
        "\n"
      ],
      "metadata": {
        "id": "fwKir-7apJQL"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "class Embedding(nn.Module):\n",
        "  def __init__(self,vocab_size: int,dim_embed: int)->None:\n",
        "    super().__init__()\n",
        "    self.embedding = nn.Embedding(vocab_size,dim_embed)\n",
        "    self.sqrt_dim_embed = mt.sqrt(dim_embed)\n",
        "  def forward(self,x:Tensor)-> Tensor:\n",
        "    x = self.embedding(x.long())\n",
        "    x = x*self.sqrt_dim_embed\n",
        "    return x\n"
      ],
      "metadata": {
        "id": "bDpG6Etx2eFc"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "import torch\n",
        "import torch.nn as nn\n",
        "\n",
        "# Define the embedding layer\n",
        "embedding = Embedding(vocab_size=4, dim_embed=3)\n",
        "\n",
        "# Example indices\n",
        "input_indices = torch.tensor([0,2,1,3,3,2,2,2,2], dtype=torch.long)  # Input tensor for indices 0 and 2\n",
        "\n",
        "# Get embeddings\n",
        "embeddings = embedding(input_indices)\n",
        "print(embeddings)\n"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Z3YnV-wt8d6Z",
        "outputId": "57e4fcb7-954c-4fc0-e2d8-d9eb91de0544"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "tensor([[-0.0535,  1.3001, -1.4354],\n",
            "        [ 2.0720,  0.0472, -1.9047],\n",
            "        [-2.0743,  1.7285,  1.0077],\n",
            "        [ 0.7708, -0.0562, -3.4704],\n",
            "        [ 0.7708, -0.0562, -3.4704],\n",
            "        [ 2.0720,  0.0472, -1.9047],\n",
            "        [ 2.0720,  0.0472, -1.9047],\n",
            "        [ 2.0720,  0.0472, -1.9047],\n",
            "        [ 2.0720,  0.0472, -1.9047]], grad_fn=<MulBackward0>)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# # atleast I am done with the embeddings\n",
        "# import math as mt\n",
        "# impor numpy as np\n",
        "# import torch\n",
        "# from torch import Tensor\n",
        "\n",
        "# pos_enc = torch.zeros(max_len,dim_embed)\n",
        "# for pos in range(max_len):\n",
        "#   for i in range(0,dim_embed,2):\n",
        "#     pos_enc[pos,i] = np.sin(pos/10000**(2*i/dim_embed))\n",
        "#     pos_enc[pos,i+1] = np.cos(pos/10000**(2*i/dim_embed))\n",
        "# return pos_enc\n",
        "\n",
        "\n",
        "# But this is not efficint because of two loops\n",
        "\n",
        "#So may be we can try smtg\n",
        "\n",
        "\n",
        "\n"
      ],
      "metadata": {
        "id": "AygXnOcO89wl"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# let us make a small model\n",
        "import torch.nn as nn\n",
        "from torch import Tensor\n",
        "class TinyModel(nn.Module):\n",
        "  def __init__(self):\n",
        "    super(TinyModel,self).__init__()\n",
        "    self.linear1 = nn.Linear(100,200)\n",
        "    self.activation = nn.ReLU()\n",
        "    self.linear2 = nn.Linear(200,10)\n",
        "    self.softmax = nn.Softmax()\n",
        "\n",
        "  def forward(self,x:Tensor)-> Tensor:\n",
        "    x = self.linear1(x)\n",
        "    x = self.activation(x)\n",
        "    x = self.linear2(x)\n",
        "    x = self.softmax(x)\n",
        "    return x\n",
        "\n",
        "\n",
        "\n"
      ],
      "metadata": {
        "id": "WwWPYkZVW6-I"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "tinymodel = TinyModel()\n",
        "input = torch.ones([1,100])\n",
        "output = tinymodel(input)\n",
        "print(output)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "J5e4cGXva2O5",
        "outputId": "c5fc6f7a-1fe9-4e89-97d4-d0eb8779a546"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "tensor([[0.1036, 0.0976, 0.0775, 0.0883, 0.1212, 0.1143, 0.1236, 0.0591, 0.0900,\n",
            "         0.1248]], grad_fn=<SoftmaxBackward0>)\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py:1511: UserWarning: Implicit dimension choice for softmax has been deprecated. Change the call to include dim=X as an argument.\n",
            "  return self._call_impl(*args, **kwargs)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# Let us do positional Encoding with classes as the two for loops might take time for many pos encodings\n",
        "class PositionEncoding(nn.Module):\n",
        "  def __init__(self,max_pos:int,dim_emb:int,drop_prob:float)->None:\n",
        "    super(PositionEncoding,self).__init()\n",
        "\n",
        "    position = torch.arange(0,max_pos).unsqueeze(1)\n",
        "    dim_pair = torch.arange(0,dim_emb,2).unsqueeze(0)\n",
        "    div_term = np.exp(dim_pair*-(mt.log(10000)/dim_emb))\n",
        "\n",
        "    pe = torch.zeros(max_pos,dim_emb)\n",
        "    pe[:,0::2] = torch.sin(position*div_term)\n",
        "    pe[:,1::2] = torch.cos(position*div_term)\n",
        "\n",
        "    pe = pe.unsqueeze(0)\n",
        "    # this adds a dimenio to the tensor and makes it in the x axis\n",
        "\n",
        "    self.register_buffer(\"pe\",pe) #making it non learnable and then storing\n",
        "\n",
        "    self.dropout = nn.Dropout(p=drop_prob)\n",
        "\n",
        "  def forward(self,x:Tensor)->Tensor:\n",
        "    max_seq_len = x.size(dim =1)\n",
        "    x = x+self.pe[:,:max_seq_len]\n",
        "    x = self.dropout(x)\n",
        "    return x"
      ],
      "metadata": {
        "id": "NvO0WC10bF6a"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# scaled dot product attention\n",
        "def attention(Q:Tensor,K:Tensor,V:Tensor,mask : Tensor = None)-> Tensor:\n",
        "  attention = np.matmul(Q,K.transpose)\n",
        "  dim_k = mt.sqrt**K.size(1)\n",
        "  attention = attention*(1/dim_k)\n",
        "  if mask is not None:\n",
        "    attention = attention.masked_fill(mask==0,1e-9)\n",
        "\n",
        "  weights = torch.nn.functional.softmax(attention,dim = -1 )\n",
        "\n",
        "\n",
        "  values = np.matmul(weights*V)\n",
        "  return values\n"
      ],
      "metadata": {
        "id": "-6-kB-gfpg_n"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "x= torch.ones(2,3)\n",
        "x.size()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "9wO2xv8hu5Ep",
        "outputId": "ce80e7db-fb98-4ae7-b7d9-24eecbd4209d"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "torch.Size([2, 3])"
            ]
          },
          "metadata": {},
          "execution_count": 14
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "print(x)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "lIdE2LWGvKF6",
        "outputId": "ae872cbd-75c0-4458-c43e-87293aefbf59"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "tensor([[1., 1., 1.],\n",
            "        [1., 1., 1.]])\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "x.transpose(0,1)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "UeIegqCWvMAs",
        "outputId": "1cb18098-7f1c-4e33-e20f-e99ed9a60497"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "tensor([[1., 1.],\n",
              "        [1., 1.],\n",
              "        [1., 1.]])"
            ]
          },
          "metadata": {},
          "execution_count": 16
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "x.transpose(-2,-1)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Xi9HXjKhvPM3",
        "outputId": "d8371053-4cd4-4aa1-c5fa-7917020829f2"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "tensor([[1., 1.],\n",
              "        [1., 1.],\n",
              "        [1., 1.]])"
            ]
          },
          "metadata": {},
          "execution_count": 17
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# let us learn about masking afterwards and go on to multihead attention\n",
        "class MultiHeadAttention(nn.Module):\n",
        "  def __init__(self,num_heads:int,dim_emb:int,drop_prob:float)->None:\n",
        "    super().__init()\n",
        "    assert dim_emb % num_heads == 0\n",
        "    self.num_heads = num_heads\n",
        "    self.dim_emb = dim_emb\n",
        "    self.dim_heads = dim_emb//num_heads\n",
        "\n",
        "    self.query = nn.Linear(dim_emb,dim_emb)\n",
        "    self.key = nn.Linear(dim_emb,dim_emb)\n",
        "    self.value = nn.Linear(dim_emb,dim_emb)\n",
        "    self.droupout = nn.Dropout(p = drop_prob)\n",
        "    self.output = nn.Linear(dim_emb,dim_emb)\n",
        "\n",
        "  def forward(self, x:Tensor, y: Tensor, mask: Tensor = None)-> Tensor:\n",
        "    query = self.query(x)\n",
        "    key = self.query(y)\n",
        "    value = self.query(y)\n",
        "\n",
        "    batch_size = x.size(0)\n",
        "\n",
        "    query = query.view(batch_size,-1,self.num_heads,self.dim_heads)\n",
        "    key   = key.view(batch_size,-1,self.num_heads,self.dim_heads)\n",
        "    value = value.view(batch_size,-1,self.num_heads,self.dim_heads)\n",
        "\n",
        "\n",
        "    query = query.transpose(1,2)\n",
        "    key = key.transpose(1,2)\n",
        "    value = value.transpose(1,2)\n",
        "\n",
        "    if mask is not None:\n",
        "      mask  = mask.unsqueeze(1)\n",
        "\n",
        "    attention = attention(query,key,value,mask = None)\n",
        "\n",
        "    attention = attention.transpose(1,2).contiguous().view(batch_size,-1,self.dim_emb)\n",
        "\n",
        "    out = self.droupout(self.output(attention))\n",
        "\n",
        "    return out"
      ],
      "metadata": {
        "id": "uAdTpufivR0x"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "class PositionwiseFeedForward(nn.Module):\n",
        "  def __init__(self,dim_emb:int,dim_pffn:int,drop_prob:float)->None:\n",
        "    super().__init()\n",
        "\n",
        "  #   self.Linear1 = nn.Linear(dim_emb,dim_pffn)\n",
        "  #   self.activation = nn.ReLU()\n",
        "  #   self.Linear2 = nn.Linear(dim_pffn,dim_emb)\n",
        "  #   self.dropout = nn.Dropout(p=drop_prob)\n",
        "\n",
        "  # def forward(self,x:Tensor)->Tensor:\n",
        "  #   x = self.Linear1(x)\n",
        "  #   x = self.activation(x)\n",
        "  #   x = self.dropout(x)\n",
        "  #   x = self.Linear2(x)\n",
        "  #   x  = self.dropout(x)\n",
        "\n",
        "    self.pffn = nn.Sequential(nn.Linear(dim_emb,dim_pffn),nn.ReLU(inplace = True),\n",
        "                              nn.Dropout(p=drop_prob),nn.Linear(dim_pffn,dim_emb)\n",
        "                              ,nn.Dropout(p=drop_prob))\n",
        "  def forward(self,x:Tensor)->Tensor:\n",
        "    x = self.pffn(x)\n",
        "    return x\n",
        "\n",
        "# these are the two methods we can use to define a model in pyTorch\n"
      ],
      "metadata": {
        "id": "l7aWrf-_OdnN"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# now we will define a encoder which will have amny encoder blocks and decoder which will have many decoder blocks\n",
        "class Encoder(nn.Module):\n",
        "  def __init__(self,num_blocks:int,num_heads:int,dim_emb:int,dim_pffn:int,drop_prob:float)->None:\n",
        "    super().__init__()\n",
        "\n",
        "    self.blocks = nn.ModuleList(\n",
        "        Encoderblock(num_heads,dim_emb,dim_pffn,drop_prob) for _ in range(num_blocks)\n",
        "    )\n",
        "    self.layernorm = nn.LayerNorm(dim_emb)\n",
        "  def forward(self,x:Tensor,x_mask = Tensor)-> Tensor:\n",
        "    for block in self.blocks:\n",
        "      x = block(x,x_mask)\n",
        "    x = x.layer_norm(x)\n",
        "    return x\n"
      ],
      "metadata": {
        "id": "3GjAwJtXQYLc"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "class Encoderblock(nn.Module):\n",
        "  def __init__(self,num_heads:int,dim_emb:int,dim_pffn:int,drop_prob:float)->None:\n",
        "    super().__init__()\n",
        "\n",
        "    self.self_attn1 = MultiHeadAttention(num_heads,dim_emb,drop_prob)\n",
        "    self.layernorm1 = nn.LayerNorm(dim_emb)\n",
        "\n",
        "    self.feedforward = PositionwiseFeedForward(dim_emb,dim_pffn,drop_prob)\n",
        "    self.layernorm2 = nn.LayerNorm(dim_emb)\n",
        "\n",
        "  def forward(self,x:Tensor,x_mask:Tensor)->Tensor:\n",
        "    x = self.layernorm1(x)\n",
        "    x = x + self.self_attn1(x,x_mask)\n",
        "    x = self.layernorm2(x)\n",
        "    x = x + self.feedforward(x)\n",
        "    return x"
      ],
      "metadata": {
        "id": "VOmUY3HeZfUH"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "class decoder(nn.Module):\n",
        "  def __init__(self,num_blocks:int,dim_emb:int,dim_pffn:int,drop_prob:float)->None:\n",
        "    super().__init__()\n",
        "    self.blocks = nn.ModuleList([Decoderblocks(dim_emb,dim_pffn,drop_prob) for _ in range(num_blocks)])\n",
        "    self.layernorm = nn.LayerNorm(dim_emb)\n",
        "\n",
        "  def forward(self,x:Tensor,x_mask:Tensor,y:Tensor,y_mask:Tensor)->Tensor:\n",
        "    for block in self.blocks:\n",
        "      y = block(y,y_mask,x,x_mask)\n",
        "    y = self.layernorm(y)\n",
        "    return y"
      ],
      "metadata": {
        "id": "BQGXf9luCb-V"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "class Decoderblocks(nn.Module):\n",
        "  def __init__(self,num_blocks:int,dim_emb:int,dim_pffn:int,drop_prob:float)->None:\n",
        "    super()._init__()\n"
      ],
      "metadata": {
        "id": "4m9lKR6YJqoZ"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "import numpy as np\n",
        "\n",
        "input = np.array([\n",
        "    [1, 2, 3, 4, 5],\n",
        "    [6, 7, 8, 9, 10],\n",
        "    [11, 12, 13, 14, 15]\n",
        "])\n"
      ],
      "metadata": {
        "id": "8le05777VMkv"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "mean_value = input.mean((-2, -1))\n",
        "print(mean_value)\n"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "5CHJzNceVNBI",
        "outputId": "4ac0bea4-26c1-48bc-921d-f520de688e99"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "8.0\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "import torch\n",
        "\n",
        "# Inputs\n",
        "x = torch.tensor([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]], dtype=torch.float32)\n",
        "\n",
        "# Layer normalization parameters\n",
        "epsilon = 1e-5\n",
        "gamma = torch.ones(x.shape[-1])\n",
        "beta = torch.zeros(x.shape[-1])"
      ],
      "metadata": {
        "id": "hA11ZR2-VOPL"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "print(gamma)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "2QPFjzBNWly7",
        "outputId": "4449bfe6-949a-4557-e13f-b3051e23dace"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "tensor([1., 1., 1.])\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "import math\n",
        "\n",
        "class MultiHeadAttention(nn.Module):\n",
        "  def __init__(self,d_model,num_heads):\n",
        "    super().__init()\n",
        "    assert d_model % num_heads == 0\n",
        "\n",
        "    self.d_model = d_model\n",
        "    self.num_heads = num_heads\n",
        "    self.d_k = self.d_model//self.num_heads\n",
        "\n",
        "    self.W_q = nn.Linear(d_model,d_model)\n",
        "    self.W_k = nn.Linear(d_model,d_model)\n",
        "    self.W_v = nn.Linear(d_model,d_model)\n",
        "    self.W_o = nn.Linear(d_model,d_model)\n",
        "\n",
        "  def scaled_dot_product_attention(self,Q,K,V,mask = None):\n",
        "\n",
        "    attn_scores = np.matmul((Q,K.transpose(-2,-1))/math.sqrt(self.d_k))\n",
        "\n",
        "    if mask is not None:\n",
        "      attn_scores = attn_scores.masked_fill_(mask == 0,1e-9)\n",
        "\n",
        "    attn_scores = torch.softmax(attn_scores,dim=-1)\n",
        "    output = np.matmul(attn_scores,V)\n",
        "\n",
        "    return output\n",
        "\n",
        "  def split_heads(self,x):\n",
        "    batch_size, seq_len,d_model = x.size()\n",
        "\n",
        "    return x.view(batch_size,seq_len,self.num_heads,self.d_k).transpose(1,2)\n",
        "\n",
        "\n",
        "  def combine_heads(self,x):\n",
        "    batch_size, _, seq_len,d_k = x.size()\n",
        "\n",
        "    return x.transpose(1,2).contiguous().view(batch_size,seq_len,self.d_model)\n",
        "\n",
        "  def forward(self,Q,K,V):\n",
        "    Q = self.split_heads(self.W_q(Q))\n",
        "    K = self.split_heads(self.W_q(K))\n",
        "    V = self.split_heads(self.W_q(V))\n",
        "\n",
        "    split_attn = self.scaled_dot_product_attention(Q,K,V,mask = None)\n",
        "\n",
        "    output = self.W_o(self.combine_heads(split_attn))\n",
        "    return output"
      ],
      "metadata": {
        "id": "Nzs_trgFWpWo"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "class PositionwiseFeedForward(nn.Module):\n",
        "  def __init__(self,d_model,d_pffn):\n",
        "    super().__init()\n",
        "    self.linear1 = nn.Linear(d_model,d_pffn)\n",
        "    self.activation = nn.ReLU()\n",
        "    self.linear2 = nn.Linear(d_pffn,d_model)\n",
        "\n",
        "  def forward(self,x):\n",
        "    x = self.linear1(x)\n",
        "    x = self.activation(x)\n",
        "    x = self.linear2(x)\n",
        "    return x"
      ],
      "metadata": {
        "id": "izziEQFnGU8d"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "class positionEncoding(nn.Module):\n",
        "  def __init__(self,max_seq_len,d_model):\n",
        "    super().__init__()\n",
        "\n",
        "    pos_enc = torch.zeros(max_seq_len,d_model)\n",
        "\n",
        "    position = torch.arange(0,max_seq_len,dtype = torch.float).unsqueeze(1)\n",
        "    div_term = torch.exp(torch.arange(0,d_model,2).float()*(-math.log(10000)/d_model))\n",
        "\n",
        "    pos_enc[:,0:2] = torch.sin(position,div_term)\n",
        "    pos_enc[:,1:2] = torch.cos(position,div_term)\n",
        "\n",
        "    self.register(\"pos_enc\",pos_enc.unsqueeze(0))\n",
        "\n",
        "  def forward(self,x):\n",
        "    x =x+self.pos_enc[:,x.size(1)]\n",
        "    return x\n"
      ],
      "metadata": {
        "id": "wySHLNFMHtHm"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "class EncoderLayer(nn.Module):\n",
        "  def __init__(self,num_heads,d_model,d_pffn,drop_prob):\n",
        "    super().__init__()\n",
        "\n",
        "    self.self_attn = MultiHeadAttention(d_model,num_heads)\n",
        "    self.feed_forward = PositionwiseFeedForward(d_model,d_pffn)\n",
        "    self.norm1 = nn.LayerNorm(d_model)\n",
        "    self.norm2 = nn.LayerNorm(d_model)\n",
        "    self.dropout = nn.Dropout(drop_prob)\n",
        "\n",
        "  def forward(self,x,mask):\n",
        "    attn_output = self.self_attn(x,x,x,mask)\n",
        "\n",
        "    x = self.norm1(x+self.dropout(attn_output))\n",
        "\n",
        "    ff_output = self.feed_forward(x)\n",
        "\n",
        "    x = self.norm2(x+self.dropout(ff_output))\n",
        "\n",
        "    return x\n"
      ],
      "metadata": {
        "id": "bdN-9b25Kw2o"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "class DecoderLayer(nn.Module):\n",
        "  def __init__(self,d_model,num_heads,d_pffn,drop_prob):\n",
        "    super().__init__()\n",
        "\n",
        "    self.self_attn = MultiHeadAttention(d_model,num_heads)\n",
        "    self.cross_attn = MultiHeadAttention(d_model,num_heads)\n",
        "    self.norm1 = nn.Linear(d_model)\n",
        "    self.norm2 = nn.Linear(d_model)\n",
        "    self.norm3 = nn.Linear(d_model)\n",
        "    self.dropout = nn.Dropout(drop_prob)\n",
        "    self.feed_forward = nn.PositionwiseFeedForward(d_model,d_pffn)\n",
        "\n",
        "  def forward(self,x_dec,enc_out,src_msk,tgt_msk):\n",
        "    attn_output = self.self_attn(x_dec,x_dec,x_dec,tgt_msk)\n",
        "\n",
        "    x = self.norm1(x+self.dropout(attn_output))\n",
        "\n",
        "    attn_output = self.cross_attn(x_dec,enc_out,enc_out,src_msk)\n",
        "\n",
        "    x = self.norm2(x+self.dropout(attn_output))\n",
        "\n",
        "    ff_output = self.feed_forward(x)\n",
        "\n",
        "    x = self.norm3(x+self.dropout(ff_output))\n",
        "    return x"
      ],
      "metadata": {
        "id": "2zw7akbCWq2y"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "class Transformer(nn.Module):\n",
        "  def __init__(self,src_vocabsize,tgt_vocabsize,num_layers,num_heads,d_model,d_pffn,max_seqlen,drop_prob):\n",
        "    super().__init__()\n",
        "\n",
        "    self.enc_embed = nn.Embedding(src_vocabsize,d_model)\n",
        "    self.dec_embed = nn.Embedding(tgt_vocabsize,d_model)\n",
        "\n",
        "    self.pos_enc = positionEncoding(max_seqlen,d_model)\n",
        "\n",
        "    self.enc_layers = nn.ModuleList([EncoderLayer(num_heads,d_model,d_pffn,drop_prob) for _ in range(num_layers)])\n",
        "    self.dec_layers = nn.ModuleList([DecoderLayer(num_heads,d_model,d_pffn,drop_prob) for _ in range(num_layers)])\n",
        "\n",
        "    self.dropout = nn.Dropout(drop_prob)\n",
        "    self.fc = nn.Linear(d_model,tgt_vocabsize)\n",
        "\n",
        "  def generate_mask(self, src, tgt):\n",
        "    src_mask = (src != 0).unsqueeze(1).unsqueeze(2)\n",
        "    tgt_mask = (tgt != 0).unsqueeze(1).unsqueeze(3)\n",
        "    seq_length = tgt.size(1)\n",
        "    nopeak_mask = (1 - torch.triu(torch.ones(1, seq_length, seq_length), diagonal=1)).bool()\n",
        "    tgt_mask = tgt_mask & nopeak_mask\n",
        "    return src_mask, tgt_mask\n",
        "\n",
        "  def forward(self,src, tgt):\n",
        "    src_mask, tgt_mask = self.generate_mask(src,tgt)\n",
        "\n",
        "    src_embedding = self.dropout(self.pos_enc(self.enc_embed(src)))\n",
        "    tgt_embedding = self.dropout(self.pos_enc(self.dec_embed(tgt)))\n",
        "\n",
        "    enc_output = src_embedding\n",
        "\n",
        "    for enc_layer in self.enc_layers:\n",
        "      enc_output = enc_layer(enc_output,src_mask)\n",
        "\n",
        "    dec_output = tgt_embedding\n",
        "\n",
        "    for dec_layer in self.dec_layers:\n",
        "      dec_output = dec_layer(dec_output,enc_output,src_mask,tgt_mask)\n",
        "\n",
        "    output = self.fc(dec_output)\n",
        "\n",
        "    return output\n",
        "\n",
        "\n",
        "\n",
        "\n",
        "\n"
      ],
      "metadata": {
        "id": "VC6EK_ZfYzfr"
      },
      "execution_count": null,
      "outputs": []
    }
  ]
}