2023-07-23 13:18:13 -05:00
{
"cells": [
2024-03-19 09:26:26 -05:00
{
"cell_type": "markdown",
"id": "f896245e-57c4-48fd-854f-9e43f22e10c9",
"metadata": {},
"source": [
2024-05-24 07:15:41 -05:00
"<table style=\"width:100%\">\n",
"<tr>\n",
"<td style=\"vertical-align:middle; text-align:left;\">\n",
"<font size=\"2\">\n",
"Supplementary code for the <a href=\"http://mng.bz/orYv\">Build a Large Language Model From Scratch</a> book by <a href=\"https://sebastianraschka.com\">Sebastian Raschka</a><br>\n",
"<br>Code repository: <a href=\"https://github.com/rasbt/LLMs-from-scratch\">https://github.com/rasbt/LLMs-from-scratch</a>\n",
"</font>\n",
"</td>\n",
"<td style=\"vertical-align:middle; text-align:left;\">\n",
"<a href=\"http://mng.bz/orYv\"><img src=\"https://sebastianraschka.com/images/LLMs-from-scratch-images/cover-small.webp\" width=\"100px\"></a>\n",
"</td>\n",
"</tr>\n",
"</table>\n"
2024-03-19 09:26:26 -05:00
]
},
2023-07-23 13:18:13 -05:00
{
"cell_type": "markdown",
"id": "ca7fc8a0-280c-4979-b0c7-fc3a99b3b785",
"metadata": {},
"source": [
2023-09-22 07:01:08 -05:00
"# Appendix A: Introduction to PyTorch (Part 1)"
2023-07-23 13:18:13 -05:00
]
},
{
"cell_type": "markdown",
"id": "f5bf13d2-8fc2-483e-88cc-6b4310221e68",
"metadata": {},
"source": [
2023-09-22 07:01:08 -05:00
"## A.1 What is PyTorch"
2023-07-23 13:18:13 -05:00
]
},
{
"cell_type": "code",
"execution_count": 1,
"id": "96ee5660-5327-48e2-9104-a882b3b2afa4",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
2024-07-24 21:53:41 -05:00
"2.4.0\n"
2023-07-23 13:18:13 -05:00
]
}
],
"source": [
"import torch\n",
"\n",
"print(torch.__version__)"
]
},
{
"cell_type": "code",
"execution_count": 2,
"id": "f73ad4e4-7ec6-4467-a9e9-0cdf6d195264",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"False\n"
]
}
],
"source": [
"print(torch.cuda.is_available())"
]
},
2024-03-29 20:42:32 -05:00
{
"cell_type": "markdown",
"id": "397ba1ab-3306-4965-8618-1ed5f24fb939",
"metadata": {},
"source": [
"<img src=\"https://sebastianraschka.com/images/LLMs-from-scratch-images/appendix-a_compressed/1.webp\" width=\"400px\">"
]
},
{
"cell_type": "markdown",
"id": "1e3c0555-88f6-4515-8c99-aa56b0769d54",
"metadata": {},
"source": [
"<img src=\"https://sebastianraschka.com/images/LLMs-from-scratch-images/appendix-a_compressed/2.webp\" width=\"300px\">\n",
"\n",
"<img src=\"https://sebastianraschka.com/images/LLMs-from-scratch-images/appendix-a_compressed/3.webp\" width=\"300px\">\n",
"\n",
"<img src=\"https://sebastianraschka.com/images/LLMs-from-scratch-images/appendix-a_compressed/4.webp\" width=\"500px\">\n",
"\n",
"<img src=\"https://sebastianraschka.com/images/LLMs-from-scratch-images/appendix-a_compressed/5.webp\" width=\"500px\">"
]
},
2023-07-23 13:18:13 -05:00
{
"cell_type": "markdown",
"id": "2100cf2e-7459-4ab3-92a8-43e86ab35a9b",
"metadata": {},
"source": [
2023-09-22 07:01:08 -05:00
"## A.2 Understanding tensors"
2023-07-23 13:18:13 -05:00
]
},
2024-03-29 20:42:32 -05:00
{
"cell_type": "markdown",
"id": "3c484e87-bfc9-4105-b0a7-1e23b2a72a30",
"metadata": {},
"source": [
"<img src=\"https://sebastianraschka.com/images/LLMs-from-scratch-images/appendix-a_compressed/6.webp\" width=\"400px\">"
]
},
2023-07-23 13:18:13 -05:00
{
"cell_type": "markdown",
"id": "26d7f785-e048-42bc-9182-a556af6bb7f4",
"metadata": {},
"source": [
2023-09-22 07:01:08 -05:00
"### A.2.1 Scalars, vectors, matrices, and tensors"
2023-07-23 13:18:13 -05:00
]
},
{
"cell_type": "code",
"execution_count": 3,
"id": "a3a464d6-cec8-4363-87bd-ea4f900baced",
"metadata": {},
"outputs": [],
"source": [
"import torch\n",
"import numpy as np\n",
"\n",
"# create a 0D tensor (scalar) from a Python integer\n",
"tensor0d = torch.tensor(1)\n",
"\n",
"# create a 1D tensor (vector) from a Python list\n",
"tensor1d = torch.tensor([1, 2, 3])\n",
"\n",
"# create a 2D tensor from a nested Python list\n",
2024-06-02 20:41:34 -05:00
"tensor2d = torch.tensor([[1, 2], \n",
" [3, 4]])\n",
2023-07-23 13:18:13 -05:00
"\n",
"# create a 3D tensor from a nested Python list\n",
2024-06-02 20:41:34 -05:00
"tensor3d_1 = torch.tensor([[[1, 2], [3, 4]], \n",
" [[5, 6], [7, 8]]])\n",
2023-07-23 13:18:13 -05:00
"\n",
"# create a 3D tensor from NumPy array\n",
2024-06-02 20:41:34 -05:00
"ary3d = np.array([[[1, 2], [3, 4]], \n",
" [[5, 6], [7, 8]]])\n",
2023-07-23 13:18:13 -05:00
"tensor3d_2 = torch.tensor(ary3d) # Copies NumPy array\n",
"tensor3d_3 = torch.from_numpy(ary3d) # Shares memory with NumPy array"
]
},
{
"cell_type": "code",
"execution_count": 4,
"id": "dbe14c47-499a-4d48-b354-a0e6fd957872",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"tensor([[[1, 2],\n",
" [3, 4]],\n",
"\n",
" [[5, 6],\n",
" [7, 8]]])\n"
]
}
],
"source": [
"ary3d[0, 0, 0] = 999\n",
"print(tensor3d_2) # remains unchanged"
]
},
{
"cell_type": "code",
"execution_count": 5,
"id": "e3e4c23a-cdba-46f5-a2dc-5fb32bf9117b",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"tensor([[[999, 2],\n",
" [ 3, 4]],\n",
"\n",
" [[ 5, 6],\n",
" [ 7, 8]]])\n"
]
}
],
"source": [
"print(tensor3d_3) # changes because of memory sharing"
]
},
{
"cell_type": "markdown",
"id": "63dec48d-2b60-41a2-ac06-fef7e718605a",
"metadata": {},
"source": [
2023-09-22 07:01:08 -05:00
"### A.2.2 Tensor data types"
2023-07-23 13:18:13 -05:00
]
},
{
"cell_type": "code",
"execution_count": 6,
"id": "3f48c014-e1a2-4a53-b5c5-125812d4034c",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"torch.int64\n"
]
}
],
"source": [
"tensor1d = torch.tensor([1, 2, 3])\n",
"print(tensor1d.dtype)"
]
},
{
"cell_type": "code",
"execution_count": 7,
"id": "5429a086-9de2-4ac7-9f14-d087a7507394",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"torch.float32\n"
]
}
],
"source": [
"floatvec = torch.tensor([1.0, 2.0, 3.0])\n",
"print(floatvec.dtype)"
]
},
{
"cell_type": "code",
"execution_count": 8,
"id": "a9a438d1-49bb-481c-8442-7cc2bb3dd4af",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"torch.float32\n"
]
}
],
"source": [
"floatvec = tensor1d.to(torch.float32)\n",
"print(floatvec.dtype)"
]
},
{
"cell_type": "markdown",
"id": "2020deb5-aa02-4524-b311-c010f4ad27ff",
"metadata": {},
"source": [
2023-09-22 07:01:08 -05:00
"### A.2.3 Common PyTorch tensor operations"
2023-07-23 13:18:13 -05:00
]
},
{
"cell_type": "code",
"execution_count": 9,
"id": "c02095f2-8a48-4953-b3c9-5313d4362ce7",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"tensor([[1, 2, 3],\n",
" [4, 5, 6]])"
]
},
"execution_count": 9,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
2024-06-02 20:41:34 -05:00
"tensor2d = torch.tensor([[1, 2, 3], \n",
" [4, 5, 6]])\n",
2023-07-23 13:18:13 -05:00
"tensor2d"
]
},
{
"cell_type": "code",
"execution_count": 10,
"id": "f33e1d45-5b2c-4afe-b4b2-66ac4099fd1a",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"torch.Size([2, 3])"
]
},
"execution_count": 10,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"tensor2d.shape"
]
},
{
"cell_type": "code",
"execution_count": 11,
"id": "f3a4129d-f870-4e03-9c32-cd8521cb83fe",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"tensor([[1, 2],\n",
" [3, 4],\n",
" [5, 6]])"
]
},
"execution_count": 11,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"tensor2d.reshape(3, 2)"
]
},
{
"cell_type": "code",
"execution_count": 12,
"id": "589ac0a7-adc7-41f3-b721-155f580e9369",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"tensor([[1, 2],\n",
" [3, 4],\n",
" [5, 6]])"
]
},
"execution_count": 12,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"tensor2d.view(3, 2)"
]
},
{
"cell_type": "code",
"execution_count": 13,
"id": "344e307f-ba5d-4f9a-a791-2c75a3d1417e",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"tensor([[1, 4],\n",
" [2, 5],\n",
" [3, 6]])"
]
},
"execution_count": 13,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"tensor2d.T"
]
},
{
"cell_type": "code",
"execution_count": 14,
"id": "19a75030-6a41-4ca8-9aae-c507ae79225c",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"tensor([[14, 32],\n",
" [32, 77]])"
]
},
"execution_count": 14,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"tensor2d.matmul(tensor2d.T)"
]
},
{
"cell_type": "code",
"execution_count": 15,
"id": "e7c950bc-d640-4203-b210-3ac8932fe4d4",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"tensor([[14, 32],\n",
" [32, 77]])"
]
},
"execution_count": 15,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"tensor2d @ tensor2d.T"
]
},
{
"cell_type": "markdown",
"id": "4c15bdeb-78e2-4870-8a4f-a9f591666f38",
"metadata": {},
"source": [
2023-09-22 07:01:08 -05:00
"## A.3 Seeing models as computation graphs"
2023-07-23 13:18:13 -05:00
]
},
2024-03-29 20:42:32 -05:00
{
"cell_type": "markdown",
"id": "0f3e16c3-07df-44b6-9106-a42fb24452a9",
"metadata": {},
"source": [
"<img src=\"https://sebastianraschka.com/images/LLMs-from-scratch-images/appendix-a_compressed/7.webp\" width=\"600px\">"
]
},
2023-07-23 13:18:13 -05:00
{
"cell_type": "code",
"execution_count": 16,
"id": "22af61e9-0443-4705-94d7-24c21add09c7",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"tensor(0.0852)\n"
]
}
],
"source": [
"import torch.nn.functional as F\n",
"\n",
"y = torch.tensor([1.0]) # true label\n",
"x1 = torch.tensor([1.1]) # input feature\n",
"w1 = torch.tensor([2.2]) # weight parameter\n",
"b = torch.tensor([0.0]) # bias unit\n",
"\n",
"z = x1 * w1 + b # net input\n",
"a = torch.sigmoid(z) # activation & output\n",
"\n",
"loss = F.binary_cross_entropy(a, y)\n",
"print(loss)"
]
},
{
"cell_type": "markdown",
"id": "f9424f26-2bac-47e7-b834-92ece802247c",
"metadata": {},
"source": [
2023-09-22 07:01:08 -05:00
"## A.4 Automatic differentiation made easy"
2023-07-23 13:18:13 -05:00
]
},
2024-03-29 20:42:32 -05:00
{
"cell_type": "markdown",
"id": "33aa2ee4-6f1d-448d-8707-67cd5278233c",
"metadata": {},
"source": [
"<img src=\"https://sebastianraschka.com/images/LLMs-from-scratch-images/appendix-a_compressed/8.webp\" width=\"600px\">"
]
},
2023-07-23 13:18:13 -05:00
{
"cell_type": "code",
"execution_count": 17,
"id": "ebf5cef7-48d6-4d2a-8ab0-0fb10bdd7d1a",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"(tensor([-0.0898]),)\n",
"(tensor([-0.0817]),)\n"
]
}
],
"source": [
"import torch.nn.functional as F\n",
"from torch.autograd import grad\n",
"\n",
"y = torch.tensor([1.0])\n",
"x1 = torch.tensor([1.1])\n",
"w1 = torch.tensor([2.2], requires_grad=True)\n",
"b = torch.tensor([0.0], requires_grad=True)\n",
"\n",
"z = x1 * w1 + b \n",
"a = torch.sigmoid(z)\n",
"\n",
"loss = F.binary_cross_entropy(a, y)\n",
"\n",
"grad_L_w1 = grad(loss, w1, retain_graph=True)\n",
"grad_L_b = grad(loss, b, retain_graph=True)\n",
"\n",
"print(grad_L_w1)\n",
"print(grad_L_b)"
]
},
{
"cell_type": "code",
"execution_count": 18,
"id": "93c5875d-f6b2-492c-b5ef-7e132f93a4e0",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"tensor([-0.0898])\n",
"tensor([-0.0817])\n"
]
}
],
"source": [
"loss.backward()\n",
"\n",
"print(w1.grad)\n",
"print(b.grad)"
]
},
{
"cell_type": "markdown",
"id": "f53bdd7d-44e6-40ab-8a5a-4eef74ef35dc",
"metadata": {},
"source": [
2023-09-22 07:01:08 -05:00
"## A.5 Implementing multilayer neural networks"
2023-07-23 13:18:13 -05:00
]
},
2024-03-29 20:42:32 -05:00
{
"cell_type": "markdown",
"id": "d6cb9787-2bc8-4379-9e8c-a3401ac63c51",
"metadata": {},
"source": [
"<img src=\"https://sebastianraschka.com/images/LLMs-from-scratch-images/appendix-a_compressed/9.webp\" width=\"500px\">"
]
},
2023-07-23 13:18:13 -05:00
{
"cell_type": "code",
"execution_count": 19,
"id": "84b749e1-7768-4cfe-94d6-a08c7feff4a1",
"metadata": {},
"outputs": [],
"source": [
"class NeuralNetwork(torch.nn.Module):\n",
" def __init__(self, num_inputs, num_outputs):\n",
" super().__init__()\n",
"\n",
" self.layers = torch.nn.Sequential(\n",
" \n",
" # 1st hidden layer\n",
" torch.nn.Linear(num_inputs, 30),\n",
" torch.nn.ReLU(),\n",
"\n",
" # 2nd hidden layer\n",
" torch.nn.Linear(30, 20),\n",
" torch.nn.ReLU(),\n",
"\n",
" # output layer\n",
" torch.nn.Linear(20, num_outputs),\n",
" )\n",
"\n",
" def forward(self, x):\n",
" logits = self.layers(x)\n",
" return logits"
]
},
{
"cell_type": "code",
"execution_count": 20,
"id": "c5b59e2e-1930-456d-93b9-f69263e3adbe",
"metadata": {},
"outputs": [],
"source": [
"model = NeuralNetwork(50, 3)"
]
},
{
"cell_type": "code",
"execution_count": 21,
"id": "39d02a21-33e7-4879-8fd2-d6309faf2f8d",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"NeuralNetwork(\n",
" (layers): Sequential(\n",
" (0): Linear(in_features=50, out_features=30, bias=True)\n",
" (1): ReLU()\n",
" (2): Linear(in_features=30, out_features=20, bias=True)\n",
" (3): ReLU()\n",
" (4): Linear(in_features=20, out_features=3, bias=True)\n",
" )\n",
")\n"
]
}
],
"source": [
"print(model)"
]
},
{
"cell_type": "code",
"execution_count": 22,
"id": "94535738-de02-4c2a-9b44-1cd186fa990a",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Total number of trainable model parameters: 2213\n"
]
}
],
"source": [
"num_params = sum(p.numel() for p in model.parameters() if p.requires_grad)\n",
"print(\"Total number of trainable model parameters:\", num_params)"
]
},
{
"cell_type": "code",
"execution_count": 23,
"id": "2c394106-ad71-4ccb-a3c9-9b60af3fa748",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Parameter containing:\n",
2024-07-24 21:53:41 -05:00
"tensor([[ 0.1182, 0.0606, -0.1292, ..., -0.1126, 0.0735, -0.0597],\n",
" [-0.0249, 0.0154, -0.0476, ..., -0.1001, -0.1288, 0.1295],\n",
" [ 0.0641, 0.0018, -0.0367, ..., -0.0990, -0.0424, -0.0043],\n",
2023-07-23 13:18:13 -05:00
" ...,\n",
2024-07-24 21:53:41 -05:00
" [ 0.0618, 0.0867, 0.1361, ..., -0.0254, 0.0399, 0.1006],\n",
" [ 0.0842, -0.0512, -0.0960, ..., -0.1091, 0.1242, -0.0428],\n",
" [ 0.0518, -0.1390, -0.0923, ..., -0.0954, -0.0668, -0.0037]],\n",
2023-07-23 13:18:13 -05:00
" requires_grad=True)\n"
]
}
],
"source": [
"print(model.layers[0].weight)"
]
},
{
"cell_type": "code",
"execution_count": 24,
"id": "b201882b-9285-4db9-bb63-43afe6a2ff9e",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Parameter containing:\n",
"tensor([[-0.0577, 0.0047, -0.0702, ..., 0.0222, 0.1260, 0.0865],\n",
" [ 0.0502, 0.0307, 0.0333, ..., 0.0951, 0.1134, -0.0297],\n",
" [ 0.1077, -0.1108, 0.0122, ..., 0.0108, -0.1049, -0.1063],\n",
" ...,\n",
" [-0.0787, 0.1259, 0.0803, ..., 0.1218, 0.1303, -0.1351],\n",
" [ 0.1359, 0.0175, -0.0673, ..., 0.0674, 0.0676, 0.1058],\n",
" [ 0.0790, 0.1343, -0.0293, ..., 0.0344, -0.0971, -0.0509]],\n",
" requires_grad=True)\n"
]
}
],
"source": [
"torch.manual_seed(123)\n",
"\n",
"model = NeuralNetwork(50, 3)\n",
"print(model.layers[0].weight)"
]
},
{
"cell_type": "code",
"execution_count": 25,
"id": "1da9a35e-44f3-460c-90fe-304519736fd6",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"torch.Size([30, 50])\n"
]
}
],
"source": [
"print(model.layers[0].weight.shape)"
]
},
{
"cell_type": "code",
"execution_count": 26,
"id": "57eadbae-90fe-43a3-a33f-c23a095ba42a",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"tensor([[-0.1262, 0.1080, -0.1792]], grad_fn=<AddmmBackward0>)\n"
]
}
],
"source": [
"torch.manual_seed(123)\n",
"\n",
"X = torch.rand((1, 50))\n",
"out = model(X)\n",
"print(out)"
]
},
{
"cell_type": "code",
"execution_count": 27,
"id": "48d720cb-ef73-4b7b-92e0-8198a072defd",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"tensor([[-0.1262, 0.1080, -0.1792]])\n"
]
}
],
"source": [
"with torch.no_grad():\n",
" out = model(X)\n",
"print(out)"
]
},
{
"cell_type": "code",
"execution_count": 28,
"id": "10df3640-83c3-4061-a74d-08f07a5cc6ac",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"tensor([[0.3113, 0.3934, 0.2952]])\n"
]
}
],
"source": [
"with torch.no_grad():\n",
2024-01-28 12:25:49 -06:00
" out = torch.softmax(model(X), dim=1)\n",
2023-07-23 13:18:13 -05:00
"print(out)"
]
},
{
"cell_type": "markdown",
"id": "19858180-0f26-43a8-b2c3-7ed40abf9f85",
"metadata": {},
"source": [
2023-09-22 07:01:08 -05:00
"## A.6 Setting up efficient data loaders"
2023-07-23 13:18:13 -05:00
]
},
2024-03-29 20:42:32 -05:00
{
"cell_type": "markdown",
"id": "0f98d8fc-5618-47a2-bc72-153818972a24",
"metadata": {},
"source": [
"<img src=\"https://sebastianraschka.com/images/LLMs-from-scratch-images/appendix-a_compressed/10.webp\" width=\"600px\">"
]
},
2023-07-23 13:18:13 -05:00
{
"cell_type": "code",
"execution_count": 29,
"id": "b9dc2745-8be8-4344-80ef-325f02cda7b7",
"metadata": {},
"outputs": [],
"source": [
"X_train = torch.tensor([\n",
" [-1.2, 3.1],\n",
" [-0.9, 2.9],\n",
" [-0.5, 2.6],\n",
" [2.3, -1.1],\n",
" [2.7, -1.5]\n",
"])\n",
"\n",
"y_train = torch.tensor([0, 0, 0, 1, 1])"
]
},
{
"cell_type": "code",
"execution_count": 30,
"id": "88283948-5fca-461a-98a1-788b6be191d5",
"metadata": {},
"outputs": [],
"source": [
"X_test = torch.tensor([\n",
" [-0.8, 2.8],\n",
" [2.6, -1.6],\n",
"])\n",
"\n",
"y_test = torch.tensor([0, 1])"
]
},
{
"cell_type": "code",
"execution_count": 31,
"id": "edf323e2-1789-41a0-8e44-f3cab16e5f5d",
"metadata": {},
"outputs": [],
"source": [
"from torch.utils.data import Dataset\n",
"\n",
"\n",
"class ToyDataset(Dataset):\n",
" def __init__(self, X, y):\n",
" self.features = X\n",
" self.labels = y\n",
"\n",
" def __getitem__(self, index):\n",
" one_x = self.features[index]\n",
" one_y = self.labels[index] \n",
" return one_x, one_y\n",
"\n",
" def __len__(self):\n",
" return self.labels.shape[0]\n",
"\n",
"train_ds = ToyDataset(X_train, y_train)\n",
"test_ds = ToyDataset(X_test, y_test)"
]
},
{
"cell_type": "code",
"execution_count": 32,
"id": "b7014705-1fdc-4f72-b892-d8db8bebc331",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"5"
]
},
"execution_count": 32,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"len(train_ds)"
]
},
{
"cell_type": "code",
"execution_count": 33,
"id": "3ec6627a-4c3f-481a-b794-d2131be95eaf",
"metadata": {},
"outputs": [],
"source": [
"from torch.utils.data import DataLoader\n",
"\n",
"torch.manual_seed(123)\n",
"\n",
"train_loader = DataLoader(\n",
" dataset=train_ds,\n",
" batch_size=2,\n",
" shuffle=True,\n",
" num_workers=0\n",
")"
]
},
{
"cell_type": "code",
"execution_count": 34,
"id": "8c9446de-5e4b-44fa-bf9a-a63e2661027e",
"metadata": {},
"outputs": [],
"source": [
"test_ds = ToyDataset(X_test, y_test)\n",
"\n",
"test_loader = DataLoader(\n",
" dataset=test_ds,\n",
" batch_size=2,\n",
" shuffle=False,\n",
" num_workers=0\n",
")"
]
},
{
"cell_type": "code",
"execution_count": 35,
"id": "99d4404c-9884-419f-979c-f659742d86ef",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Batch 1: tensor([[ 2.3000, -1.1000],\n",
" [-0.9000, 2.9000]]) tensor([1, 0])\n",
"Batch 2: tensor([[-1.2000, 3.1000],\n",
" [-0.5000, 2.6000]]) tensor([0, 0])\n",
"Batch 3: tensor([[ 2.7000, -1.5000]]) tensor([1])\n"
]
}
],
"source": [
"for idx, (x, y) in enumerate(train_loader):\n",
" print(f\"Batch {idx+1}:\", x, y)"
]
},
{
"cell_type": "code",
"execution_count": 36,
"id": "9d003f7e-7a80-40bf-a7fb-7a0d7dbba9db",
"metadata": {},
"outputs": [],
"source": [
"train_loader = DataLoader(\n",
" dataset=train_ds,\n",
" batch_size=2,\n",
" shuffle=True,\n",
" num_workers=0,\n",
" drop_last=True\n",
")"
]
},
{
"cell_type": "code",
2024-03-26 08:38:27 -05:00
"execution_count": 37,
2023-07-23 13:18:13 -05:00
"id": "4db4d7f4-82da-44a4-b94e-ee04665d9c3c",
"metadata": {},
2024-03-26 08:38:27 -05:00
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Batch 1: tensor([[-1.2000, 3.1000],\n",
" [-0.5000, 2.6000]]) tensor([0, 0])\n",
"Batch 2: tensor([[ 2.3000, -1.1000],\n",
" [-0.9000, 2.9000]]) tensor([1, 0])\n"
]
}
],
2023-07-23 13:18:13 -05:00
"source": [
"for idx, (x, y) in enumerate(train_loader):\n",
" print(f\"Batch {idx+1}:\", x, y)"
]
},
2024-03-29 20:42:32 -05:00
{
"cell_type": "markdown",
"id": "eb03ed57-df38-4ee0-a553-0863450df39b",
"metadata": {},
"source": [
"<img src=\"https://sebastianraschka.com/images/LLMs-from-scratch-images/appendix-a_compressed/11.webp\" width=\"600px\">"
]
},
2023-07-23 13:18:13 -05:00
{
"cell_type": "markdown",
"id": "d904ca82-e50f-4f3d-a3ac-fc6ca53dd00e",
"metadata": {},
"source": [
2023-09-22 07:01:08 -05:00
"## A.7 A typical training loop"
2023-07-23 13:18:13 -05:00
]
},
{
"cell_type": "code",
"execution_count": 38,
"id": "93f1791a-d887-4fc5-a307-5e5bde9e06f6",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Epoch: 001/003 | Batch 000/002 | Train/Val Loss: 0.75\n",
"Epoch: 001/003 | Batch 001/002 | Train/Val Loss: 0.65\n",
"Epoch: 002/003 | Batch 000/002 | Train/Val Loss: 0.44\n",
"Epoch: 002/003 | Batch 001/002 | Train/Val Loss: 0.13\n",
"Epoch: 003/003 | Batch 000/002 | Train/Val Loss: 0.03\n",
"Epoch: 003/003 | Batch 001/002 | Train/Val Loss: 0.00\n"
]
}
],
"source": [
"import torch.nn.functional as F\n",
"\n",
"\n",
"torch.manual_seed(123)\n",
"model = NeuralNetwork(num_inputs=2, num_outputs=2)\n",
"optimizer = torch.optim.SGD(model.parameters(), lr=0.5)\n",
"\n",
"num_epochs = 3\n",
"\n",
"for epoch in range(num_epochs):\n",
" \n",
" model.train()\n",
" for batch_idx, (features, labels) in enumerate(train_loader):\n",
"\n",
" logits = model(features)\n",
" \n",
" loss = F.cross_entropy(logits, labels) # Loss function\n",
" \n",
" optimizer.zero_grad()\n",
" loss.backward()\n",
" optimizer.step()\n",
" \n",
" ### LOGGING\n",
" print(f\"Epoch: {epoch+1:03d}/{num_epochs:03d}\"\n",
" f\" | Batch {batch_idx:03d}/{len(train_loader):03d}\"\n",
" f\" | Train/Val Loss: {loss:.2f}\")\n",
"\n",
" model.eval()\n",
" # Optional model evaluation"
]
},
{
"cell_type": "code",
"execution_count": 39,
"id": "00dcf57f-6a7e-4af7-aa5a-df2cb0866fa5",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"tensor([[ 2.8569, -4.1618],\n",
" [ 2.5382, -3.7548],\n",
" [ 2.0944, -3.1820],\n",
" [-1.4814, 1.4816],\n",
" [-1.7176, 1.7342]])\n"
]
}
],
"source": [
"model.eval()\n",
"\n",
"with torch.no_grad():\n",
" outputs = model(X_train)\n",
"\n",
"print(outputs)"
]
},
{
"cell_type": "code",
"execution_count": 40,
"id": "19be7390-18b8-43f9-9841-d7fb1919f6fd",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"tensor([[ 0.9991, 0.0009],\n",
" [ 0.9982, 0.0018],\n",
" [ 0.9949, 0.0051],\n",
" [ 0.0491, 0.9509],\n",
" [ 0.0307, 0.9693]])\n",
"tensor([0, 0, 0, 1, 1])\n"
]
}
],
"source": [
"torch.set_printoptions(sci_mode=False)\n",
2024-01-28 12:25:49 -06:00
"probas = torch.softmax(outputs, dim=1)\n",
2023-07-23 13:18:13 -05:00
"print(probas)\n",
"\n",
2024-03-26 08:38:27 -05:00
"predictions = torch.argmax(probas, dim=1)\n",
2023-07-23 13:18:13 -05:00
"print(predictions)"
]
},
{
"cell_type": "code",
"execution_count": 41,
"id": "07e7e530-f8d3-429c-9f5e-cf8078078c0e",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"tensor([0, 0, 0, 1, 1])\n"
]
}
],
"source": [
"predictions = torch.argmax(outputs, dim=1)\n",
"print(predictions)"
]
},
{
"cell_type": "code",
"execution_count": 42,
"id": "5f756f0d-63c8-41b5-a5d8-01baa847e026",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"tensor([True, True, True, True, True])"
]
},
"execution_count": 42,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"predictions == y_train"
]
},
{
"cell_type": "code",
"execution_count": 43,
"id": "da274bb0-f11c-4c81-a880-7a031fbf2943",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"tensor(5)"
]
},
"execution_count": 43,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"torch.sum(predictions == y_train)"
]
},
{
"cell_type": "code",
"execution_count": 44,
"id": "16d62314-8dee-45b0-8f55-9e5aae2b24f4",
"metadata": {},
"outputs": [],
"source": [
"def compute_accuracy(model, dataloader):\n",
"\n",
" model = model.eval()\n",
" correct = 0.0\n",
" total_examples = 0\n",
" \n",
" for idx, (features, labels) in enumerate(dataloader):\n",
" \n",
" with torch.no_grad():\n",
" logits = model(features)\n",
" \n",
" predictions = torch.argmax(logits, dim=1)\n",
" compare = labels == predictions\n",
" correct += torch.sum(compare)\n",
" total_examples += len(compare)\n",
"\n",
" return (correct / total_examples).item()"
]
},
{
"cell_type": "code",
"execution_count": 45,
"id": "4f6c9c17-2a5f-46c0-804b-873f169b729a",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"1.0"
]
},
"execution_count": 45,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"compute_accuracy(model, train_loader)"
]
},
{
"cell_type": "code",
"execution_count": 46,
"id": "311ed864-e21e-4aac-97c7-c6086caef27a",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"1.0"
]
},
"execution_count": 46,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"compute_accuracy(model, test_loader)"
]
},
{
"cell_type": "markdown",
"id": "4d5cd469-3a45-4394-944b-3ce543f41dac",
"metadata": {},
"source": [
2023-09-22 07:01:08 -05:00
"## A.8 Saving and loading models"
2023-07-23 13:18:13 -05:00
]
},
{
"cell_type": "code",
"execution_count": 47,
"id": "b013127d-a2c3-4b04-9fb3-a6a7c88d83c5",
"metadata": {},
"outputs": [],
"source": [
"torch.save(model.state_dict(), \"model.pth\")"
]
},
{
"cell_type": "code",
"execution_count": 48,
"id": "b2b428c2-3a44-4d91-97c4-8298cf2b51eb",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"<All keys matched successfully>"
]
},
"execution_count": 48,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
2024-06-26 00:30:30 +02:00
"model = NeuralNetwork(2, 2) # needs to match the original model exactly\n",
2024-07-24 21:53:41 -05:00
"model.load_state_dict(torch.load(\"model.pth\", weights_only=True))"
2023-07-23 13:18:13 -05:00
]
},
{
"cell_type": "markdown",
"id": "f891c013-43da-4a05-973d-997be313d2d8",
"metadata": {},
"source": [
2023-09-22 07:01:08 -05:00
"## A.9 Optimizing training performance with GPUs"
2023-07-23 13:18:13 -05:00
]
},
{
"cell_type": "markdown",
"id": "e68ae888-cabf-49c9-bad6-ecdce774db57",
"metadata": {},
"source": [
2023-09-22 07:01:08 -05:00
"### A.9.1 PyTorch computations on GPU devices"
2023-07-23 13:18:13 -05:00
]
},
{
"cell_type": "markdown",
"id": "141c845f-efe3-4614-b376-b8b7a9a2c887",
"metadata": {},
"source": [
2023-09-22 07:01:08 -05:00
"See [code-part2.ipynb](code-part2.ipynb)"
2023-07-23 13:18:13 -05:00
]
},
{
"cell_type": "markdown",
"id": "99811829-b817-42ea-b03e-d35374debcc0",
"metadata": {},
"source": [
2023-09-22 07:01:08 -05:00
"### A.9.2 Single-GPU training"
2023-07-23 13:18:13 -05:00
]
},
{
"cell_type": "markdown",
"id": "0b21456c-4af7-440f-9e78-37770277b5bc",
"metadata": {},
"source": [
2023-09-22 07:01:08 -05:00
"See [code-part2.ipynb](code-part2.ipynb)"
2023-07-23 13:18:13 -05:00
]
},
{
"cell_type": "markdown",
"id": "db6eb2d1-a341-4489-b04b-635c26945333",
"metadata": {},
"source": [
2023-09-22 07:01:08 -05:00
"### A.9.3 Training with multiple GPUs"
2023-07-23 13:18:13 -05:00
]
},
{
"cell_type": "markdown",
"id": "9d049a81-5fb0-49b5-9d6a-17a9976d8520",
"metadata": {},
"source": [
2024-01-07 15:13:53 -06:00
"See [DDP-script.py](DDP-script.py)"
2023-07-23 13:18:13 -05:00
]
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"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",
2024-07-24 21:53:41 -05:00
"version": "3.11.4"
2023-07-23 13:18:13 -05:00
}
},
"nbformat": 4,
"nbformat_minor": 5
}
