fashion-mnist/notebooks/fashion-mnist.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "from torch import nn\n",
    "from torch.utils.data import DataLoader\n",
    "from torchvision import datasets\n",
    "from torchvision.transforms import ToTensor\n",
    "from torchvision.utils import save_image\n",
    "import matplotlib.pyplot as plt\n",
    "import gzip"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Download training data from open datasets.\n",
    "training_data = datasets.FashionMNIST(\n",
    "    root=\"data\",\n",
    "    train=True,\n",
    "    download=True,\n",
    "    transform=ToTensor(),\n",
    ")\n",
    "\n",
    "# Download test data from open datasets.\n",
    "test_data = datasets.FashionMNIST(\n",
    "    root=\"data\",\n",
    "    train=False,\n",
    "    download=True,\n",
    "    transform=ToTensor(),\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "classes = [\n",
    "    \"T-shirt/top\",\n",
    "    \"Trouser\",\n",
    "    \"Pullover\",\n",
    "    \"Dress\",\n",
    "    \"Coat\",\n",
    "    \"Sandal\",\n",
    "    \"Shirt\",\n",
    "    \"Sneaker\",\n",
    "    \"Bag\",\n",
    "    \"Ankle boot\",\n",
    "]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 800x800 with 9 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "figure = plt.figure(figsize=(8, 8))\n",
    "cols, rows = 3, 3\n",
    "for i in range(1, cols * rows + 1):\n",
    "    sample_idx = torch.randint(len(training_data), size=(1,)).item()\n",
    "    img, label = training_data[sample_idx]\n",
    "    figure.add_subplot(rows, cols, i)\n",
    "    plt.title(classes[label])\n",
    "    plt.axis(\"off\")\n",
    "    plt.imshow(img.squeeze(), cmap=\"gray\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Shape of X [N, C, H, W]: torch.Size([64, 1, 28, 28])\n",
      "Shape of y: torch.Size([64]) torch.int64\n"
     ]
    }
   ],
   "source": [
    "batch_size = 64\n",
    "\n",
    "# Create data loaders.\n",
    "train_dataloader = DataLoader(training_data, batch_size=batch_size)\n",
    "test_dataloader = DataLoader(test_data, batch_size=batch_size)\n",
    "\n",
    "for X, y in test_dataloader:\n",
    "    print(f\"Shape of X [N, C, H, W]: {X.shape}\")\n",
    "    print(f\"Shape of y: {y.shape} {y.dtype}\")\n",
    "    break"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Using cuda device\n",
      "NeuralNetwork(\n",
      "  (flatten): Flatten(start_dim=1, end_dim=-1)\n",
      "  (linear_relu_stack): Sequential(\n",
      "    (0): Linear(in_features=784, out_features=512, bias=True)\n",
      "    (1): ReLU()\n",
      "    (2): Linear(in_features=512, out_features=512, bias=True)\n",
      "    (3): ReLU()\n",
      "    (4): Linear(in_features=512, out_features=10, bias=True)\n",
      "  )\n",
      ")\n"
     ]
    }
   ],
   "source": [
    "# Get cpu, gpu or mps device for training.\n",
    "device = (\n",
    "    \"cuda\"\n",
    "    if torch.cuda.is_available()\n",
    "    else \"cpu\"\n",
    ")\n",
    "print(f\"Using {device} device\")\n",
    "\n",
    "# Define model\n",
    "class NeuralNetwork(nn.Module):\n",
    "    def __init__(self):\n",
    "        super().__init__()\n",
    "        self.flatten = nn.Flatten()\n",
    "        self.linear_relu_stack = nn.Sequential(\n",
    "            nn.Linear(28*28, 512),\n",
    "            nn.ReLU(),\n",
    "            nn.Linear(512, 512),\n",
    "            nn.ReLU(),\n",
    "            nn.Linear(512, 10)\n",
    "        )\n",
    "\n",
    "    def forward(self, x):\n",
    "        x = self.flatten(x)\n",
    "        logits = self.linear_relu_stack(x)\n",
    "        return logits\n",
    "\n",
    "model = NeuralNetwork().to(device)\n",
    "print(model)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [],
   "source": [
    "loss_fn = nn.CrossEntropyLoss()\n",
    "optimizer = torch.optim.SGD(model.parameters(), lr=1e-3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [],
   "source": [
    "def train(dataloader, model, loss_fn, optimizer):\n",
    "    size = len(dataloader.dataset)\n",
    "    model.train()\n",
    "    for batch, (X, y) in enumerate(dataloader):\n",
    "        X, y = X.to(device), y.to(device)\n",
    "\n",
    "        # Compute prediction error\n",
    "        pred = model(X)\n",
    "        loss = loss_fn(pred, y)\n",
    "\n",
    "        # Backpropagation\n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "        optimizer.zero_grad()\n",
    "\n",
    "        if batch % 100 == 0:\n",
    "            loss, current = loss.item(), (batch + 1) * len(X)\n",
    "            print(f\"loss: {loss:>7f}  [{current:>5d}/{size:>5d}]\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [],
   "source": [
    "def test(dataloader, model, loss_fn):\n",
    "    size = len(dataloader.dataset)\n",
    "    num_batches = len(dataloader)\n",
    "    model.eval()\n",
    "    test_loss, correct = 0, 0\n",
    "    with torch.no_grad():\n",
    "        for X, y in dataloader:\n",
    "            X, y = X.to(device), y.to(device)\n",
    "            pred = model(X)\n",
    "            test_loss += loss_fn(pred, y).item()\n",
    "            correct += (pred.argmax(1) == y).type(torch.float).sum().item()\n",
    "    test_loss /= num_batches\n",
    "    correct /= size\n",
    "    print(f\"Test Error: \\n Accuracy: {(100*correct):>0.1f}%, Avg loss: {test_loss:>8f} \\n\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1\n",
      "-------------------------------\n",
      "loss: 2.306957  [   64/60000]\n",
      "loss: 2.296147  [ 6464/60000]\n",
      "loss: 2.273746  [12864/60000]\n",
      "loss: 2.266272  [19264/60000]\n",
      "loss: 2.250971  [25664/60000]\n",
      "loss: 2.217182  [32064/60000]\n",
      "loss: 2.228909  [38464/60000]\n",
      "loss: 2.195453  [44864/60000]\n",
      "loss: 2.188948  [51264/60000]\n",
      "loss: 2.160021  [57664/60000]\n",
      "Test Error: \n",
      " Accuracy: 42.7%, Avg loss: 2.155762 \n",
      "\n",
      "Epoch 2\n",
      "-------------------------------\n",
      "loss: 2.167244  [   64/60000]\n",
      "loss: 2.161547  [ 6464/60000]\n",
      "loss: 2.097712  [12864/60000]\n",
      "loss: 2.116953  [19264/60000]\n",
      "loss: 2.079131  [25664/60000]\n",
      "loss: 2.011789  [32064/60000]\n",
      "loss: 2.039007  [38464/60000]\n",
      "loss: 1.964299  [44864/60000]\n",
      "loss: 1.955008  [51264/60000]\n",
      "loss: 1.901619  [57664/60000]\n",
      "Test Error: \n",
      " Accuracy: 59.4%, Avg loss: 1.893165 \n",
      "\n",
      "Epoch 3\n",
      "-------------------------------\n",
      "loss: 1.921449  [   64/60000]\n",
      "loss: 1.903921  [ 6464/60000]\n",
      "loss: 1.772288  [12864/60000]\n",
      "loss: 1.819563  [19264/60000]\n",
      "loss: 1.727786  [25664/60000]\n",
      "loss: 1.663390  [32064/60000]\n",
      "loss: 1.675190  [38464/60000]\n",
      "loss: 1.578112  [44864/60000]\n",
      "loss: 1.594267  [51264/60000]\n",
      "loss: 1.500766  [57664/60000]\n",
      "Test Error: \n",
      " Accuracy: 61.5%, Avg loss: 1.516219 \n",
      "\n",
      "Epoch 4\n",
      "-------------------------------\n",
      "loss: 1.576049  [   64/60000]\n",
      "loss: 1.555607  [ 6464/60000]\n",
      "loss: 1.387264  [12864/60000]\n",
      "loss: 1.479879  [19264/60000]\n",
      "loss: 1.362362  [25664/60000]\n",
      "loss: 1.343265  [32064/60000]\n",
      "loss: 1.353873  [38464/60000]\n",
      "loss: 1.277868  [44864/60000]\n",
      "loss: 1.322733  [51264/60000]\n",
      "loss: 1.226664  [57664/60000]\n",
      "Test Error: \n",
      " Accuracy: 62.7%, Avg loss: 1.251583 \n",
      "\n",
      "Epoch 5\n",
      "-------------------------------\n",
      "loss: 1.326068  [   64/60000]\n",
      "loss: 1.316462  [ 6464/60000]\n",
      "loss: 1.136512  [12864/60000]\n",
      "loss: 1.262430  [19264/60000]\n",
      "loss: 1.135502  [25664/60000]\n",
      "loss: 1.150901  [32064/60000]\n",
      "loss: 1.165919  [38464/60000]\n",
      "loss: 1.102995  [44864/60000]\n",
      "loss: 1.156760  [51264/60000]\n",
      "loss: 1.073268  [57664/60000]\n",
      "Test Error: \n",
      " Accuracy: 64.2%, Avg loss: 1.091607 \n",
      "\n",
      "Done!\n"
     ]
    }
   ],
   "source": [
    "epochs = 5\n",
    "for t in range(epochs):\n",
    "    print(f\"Epoch {t+1}\\n-------------------------------\")\n",
    "    train(train_dataloader, model, loss_fn, optimizer)\n",
    "    test(test_dataloader, model, loss_fn)\n",
    "print(\"Done!\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Saved PyTorch Model State to ../models/fashion-mnist-cpu.pth\n"
     ]
    }
   ],
   "source": [
    "torch.save(model.state_dict(), f\"../models/fashion-mnist-{device}.pth\")\n",
    "print(f\"Saved PyTorch Model State to ../models/fashion-mnist-{device}.pth\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<All keys matched successfully>"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model = NeuralNetwork().to(device)\n",
    "model.load_state_dict(torch.load(f\"../models/fashion-mnist-{device}.pth\"))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Predicted: \"Ankle boot\", Actual: \"Ankle boot\"\n"
     ]
    }
   ],
   "source": [
    "model.eval()\n",
    "x, y = test_data[0][0], test_data[0][1]\n",
    "\n",
    "plt.imshow(x.squeeze(), cmap=\"gray\")\n",
    "plt.show()\n",
    "\n",
    "with torch.no_grad():\n",
    "    x = x.to(device)\n",
    "    pred = model(x)\n",
    "    predicted, actual = classes[pred[0].argmax(0)], classes[y]\n",
    "    print(f'Predicted: \"{predicted}\", Actual: \"{actual}\"')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "save_image(x, \"../data/0.png\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Shape of input image: torch.Size([1, 28, 28])\n",
      "Input image center: tensor([[[0.0000, 0.0000, 0.4510, 0.4471],\n",
      "         [0.0000, 0.3490, 0.5451, 0.3529],\n",
      "         [0.3843, 0.5333, 0.4314, 0.4275],\n",
      "         [0.4588, 0.3882, 0.4353, 0.4588]]], device='cuda:0')\n"
     ]
    }
   ],
   "source": [
    "print(f\"Shape of input image: {x.shape}\")\n",
    "print(f\"Input image center: {x[:,12:16,12:16]}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Shape of flat image: torch.Size([1, 784])\n",
      "Flat image center: tensor([[0.0000, 0.0000, 0.4510, 0.4471],\n",
      "        [0.0000, 0.3490, 0.5451, 0.3529],\n",
      "        [0.3843, 0.5333, 0.4314, 0.4275],\n",
      "        [0.4588, 0.3882, 0.4353, 0.4588]], device='cuda:0')\n"
     ]
    }
   ],
   "source": [
    "flat_image = model.flatten(x)\n",
    "flat_image_center = torch.stack((\n",
    "    flat_image[:,348:352][0],\n",
    "    flat_image[:,376:380][0],\n",
    "    flat_image[:,404:408][0],\n",
    "    flat_image[:,432:436][0]))\n",
    "print(f\"Shape of flat image: {flat_image.shape}\")\n",
    "print(f\"Flat image center: {flat_image_center}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "layer 0 weight - Shape: torch.Size([512, 784]) - Parameter containing:\n",
      "tensor([[ 0.0166, -0.0215, -0.0213,  ...,  0.0249, -0.0333,  0.0195],\n",
      "        [ 0.0019,  0.0064, -0.0097,  ..., -0.0314,  0.0137, -0.0085],\n",
      "        [-0.0215, -0.0111, -0.0079,  ...,  0.0136, -0.0033,  0.0346],\n",
      "        ...,\n",
      "        [ 0.0033,  0.0057, -0.0331,  ...,  0.0188,  0.0275, -0.0074],\n",
      "        [-0.0124, -0.0270, -0.0229,  ..., -0.0219, -0.0196, -0.0023],\n",
      "        [-0.0082,  0.0086,  0.0155,  ...,  0.0073, -0.0347, -0.0143]],\n",
      "       device='cuda:0', requires_grad=True)\n",
      "\n",
      "layer 0 bias - Shape: torch.Size([512]) - Parameter containing:\n",
      "tensor([ 2.6567e-02,  6.3177e-03,  6.0578e-03,  3.2456e-03,  4.2460e-02,\n",
      "         2.5497e-03, -7.8638e-03, -1.1674e-03,  2.7715e-02,  1.0238e-02,\n",
      "        -2.8850e-02,  2.2727e-02,  8.7578e-03, -1.7382e-02,  2.3547e-02,\n",
      "        -1.5281e-02,  2.9908e-02,  8.9176e-03, -5.5175e-03,  2.0215e-02,\n",
      "         1.0769e-02,  2.2023e-02,  3.1302e-02, -1.2984e-02, -1.3096e-02,\n",
      "         6.9134e-03, -7.1418e-03, -7.6192e-03, -2.4171e-03, -1.8061e-02,\n",
      "         2.7449e-02, -1.5280e-02, -9.7236e-03,  2.5323e-02,  7.1208e-03,\n",
      "         7.8442e-03,  1.8075e-02,  3.3107e-03, -1.3616e-02, -1.4678e-02,\n",
      "        -7.4410e-03, -1.9298e-02,  1.6895e-02,  3.7959e-02,  5.6057e-03,\n",
      "         3.3199e-02,  9.0810e-03, -2.0296e-02,  3.1224e-02, -3.0898e-02,\n",
      "         3.6715e-02, -7.0851e-03,  3.3297e-02, -1.0822e-02, -2.0956e-02,\n",
      "         6.2544e-03, -2.4402e-02,  2.4104e-02,  3.0551e-02,  1.6966e-02,\n",
      "        -2.3713e-02, -9.4298e-03, -2.4084e-02, -3.6627e-02, -2.6211e-02,\n",
      "         2.2874e-02, -3.4174e-02,  3.5841e-02, -1.8281e-02, -3.1548e-02,\n",
      "         7.9167e-03, -3.2849e-02,  1.6330e-02,  2.5600e-02, -2.5716e-02,\n",
      "        -1.9726e-02, -4.3087e-03,  1.7966e-02,  1.4692e-02, -1.5646e-02,\n",
      "         3.8021e-02, -1.8460e-02, -3.9911e-03, -3.7429e-03,  3.8476e-02,\n",
      "        -8.1342e-03,  2.5503e-02, -1.3025e-02,  2.6845e-02, -1.4553e-02,\n",
      "         1.4767e-02, -2.3690e-03, -9.0481e-03, -1.2062e-02,  9.6579e-03,\n",
      "        -3.4259e-02,  3.2946e-03,  4.2605e-02, -1.3192e-02, -3.0719e-02,\n",
      "        -9.5193e-03,  9.3302e-03, -8.5024e-03,  3.2284e-02,  2.6598e-02,\n",
      "        -2.6825e-02,  6.0481e-04,  2.4803e-02, -3.3527e-02, -7.2194e-03,\n",
      "        -1.7405e-02,  3.2221e-02,  1.6279e-02, -3.3678e-02,  1.1668e-02,\n",
      "        -9.3008e-03,  2.1591e-02,  2.0855e-02,  5.2747e-03, -5.4519e-03,\n",
      "         1.7753e-02,  3.8069e-02, -2.2430e-02, -1.4328e-02, -2.5262e-02,\n",
      "         2.6640e-02, -1.3730e-02,  2.9503e-02, -5.7709e-03, -8.7418e-03,\n",
      "         2.2294e-04,  4.0905e-03,  1.1482e-02,  3.6768e-02, -1.5494e-02,\n",
      "         1.0991e-02, -2.5899e-02, -1.7274e-02, -1.7100e-02, -2.2061e-02,\n",
      "         1.2281e-02, -2.1771e-02, -2.1233e-02, -1.8577e-02, -1.3225e-02,\n",
      "         1.3118e-02, -1.3647e-02, -2.0261e-02, -2.9269e-02, -7.4688e-03,\n",
      "         2.5643e-02,  5.6130e-03, -2.3943e-02,  1.6801e-02, -4.2405e-02,\n",
      "        -2.3050e-02,  2.5098e-02, -1.2361e-02, -3.0937e-02, -1.3155e-02,\n",
      "        -9.9493e-03, -1.8265e-02, -1.5983e-03, -8.0820e-03,  2.7735e-02,\n",
      "        -2.4306e-02,  1.8375e-02,  7.6029e-03,  1.0208e-02, -5.0578e-03,\n",
      "        -4.4377e-03, -2.2246e-02, -1.6575e-03, -3.6095e-02, -2.2505e-02,\n",
      "         8.9893e-03, -6.5005e-04,  2.6137e-03, -3.2898e-02,  5.2349e-03,\n",
      "        -3.6952e-02, -1.2645e-02,  2.5605e-02, -1.8313e-02,  1.1021e-02,\n",
      "        -1.2325e-03, -8.9689e-03, -7.7161e-03,  9.4579e-03,  1.1138e-02,\n",
      "        -9.9154e-03,  6.2359e-03, -1.9614e-02,  7.7118e-04,  2.5344e-02,\n",
      "         1.8870e-02,  3.1405e-02,  2.9802e-02,  3.8646e-02, -2.6787e-02,\n",
      "        -1.7942e-03, -2.3943e-03, -2.8468e-03, -1.8455e-02,  2.2635e-02,\n",
      "        -1.3195e-02, -1.7478e-02, -1.2483e-02, -1.4699e-02,  2.8617e-02,\n",
      "        -1.9424e-02, -1.5695e-02,  2.2428e-02,  3.1033e-02, -2.8708e-02,\n",
      "         2.9764e-02,  2.4162e-02, -3.6482e-03,  3.5550e-02,  2.2663e-02,\n",
      "        -1.4393e-02, -1.9129e-02,  1.7096e-02, -2.2768e-02,  9.6114e-03,\n",
      "        -2.0193e-02, -3.1942e-02,  1.8705e-02,  2.8240e-02, -2.0161e-02,\n",
      "        -1.7300e-02,  4.7356e-03,  2.6291e-02,  1.2056e-02,  1.3331e-03,\n",
      "         1.1722e-02, -3.5914e-03, -3.0470e-02,  2.5896e-02,  1.2704e-02,\n",
      "         9.6924e-03,  5.6975e-03, -2.5815e-03, -2.6893e-02,  6.2583e-04,\n",
      "        -1.4761e-04,  3.1720e-02,  2.2210e-02, -3.1059e-02, -2.8656e-02,\n",
      "         3.6886e-02, -2.9338e-02, -3.0836e-03,  9.9971e-04, -2.0306e-03,\n",
      "        -1.5769e-02, -1.5968e-02, -1.6124e-02,  3.6038e-02, -1.2940e-02,\n",
      "        -3.0787e-02,  2.6470e-02,  2.0913e-02,  4.9916e-03,  1.6251e-02,\n",
      "        -3.6703e-02, -2.1457e-02, -1.0512e-02, -6.7004e-03,  2.0382e-02,\n",
      "         1.0368e-02,  2.4080e-02, -3.4223e-02,  1.6354e-02, -4.8280e-03,\n",
      "         3.3818e-02,  3.2567e-03,  2.8671e-02,  4.3069e-03, -3.2927e-02,\n",
      "         7.4174e-03, -1.1224e-02, -9.6772e-03, -2.8242e-03,  6.9502e-03,\n",
      "         2.9889e-02, -3.1844e-02,  5.5799e-03,  2.3559e-02,  7.8116e-03,\n",
      "        -2.3944e-03, -4.4800e-03,  2.5878e-02, -1.9889e-02,  2.8693e-02,\n",
      "        -1.3047e-02,  3.8299e-02,  2.5979e-02, -2.9567e-03, -1.1190e-02,\n",
      "         1.9961e-02, -2.4154e-02, -3.3357e-02, -6.0771e-03,  2.1705e-02,\n",
      "        -3.5205e-02, -9.8776e-03, -9.7291e-04, -3.0977e-02, -1.8364e-02,\n",
      "         3.8274e-02,  5.8001e-03, -1.3830e-03,  8.8052e-03,  2.0492e-02,\n",
      "         2.4777e-02,  7.4669e-03, -1.2838e-02, -4.2979e-03,  9.5570e-03,\n",
      "        -2.5242e-02, -2.6745e-02, -2.7924e-02,  3.4839e-02, -2.0664e-02,\n",
      "         4.1514e-03,  1.0210e-02,  1.2371e-02, -1.7355e-02, -1.0443e-02,\n",
      "         9.9498e-03,  2.9435e-02,  3.5672e-02, -2.2971e-02,  8.0021e-05,\n",
      "        -2.2618e-02, -2.6618e-02,  2.6940e-02, -1.0573e-02,  7.0781e-03,\n",
      "        -8.9479e-04,  2.0772e-02, -1.9597e-02,  3.5028e-02,  2.3409e-02,\n",
      "        -4.8420e-03, -2.3832e-02,  3.6551e-03, -1.8476e-02, -1.6016e-02,\n",
      "         1.1032e-02, -3.7018e-02,  2.7842e-02, -2.7090e-02,  5.6286e-03,\n",
      "        -5.2984e-03,  1.7749e-03, -1.1543e-02,  1.6415e-02, -5.9608e-03,\n",
      "        -3.2863e-02,  1.6371e-02,  2.9936e-02,  2.4401e-02,  4.2837e-02,\n",
      "        -3.2157e-02,  6.6676e-03, -6.8194e-03, -2.9928e-02, -3.2264e-02,\n",
      "        -1.6400e-02, -1.5021e-02, -1.5593e-02,  3.7694e-02, -9.1247e-03,\n",
      "         6.7357e-03, -3.4188e-02,  3.2778e-02,  9.3534e-03, -1.7840e-02,\n",
      "        -2.7298e-02,  4.3024e-02, -6.1444e-03,  3.7526e-02,  2.2065e-02,\n",
      "         1.8971e-02,  1.6219e-02,  2.7179e-02,  2.1602e-02,  3.5208e-02,\n",
      "        -1.6234e-02,  9.4198e-05,  3.2684e-02, -2.2394e-02, -3.3205e-02,\n",
      "         1.1710e-02,  1.8493e-02, -3.5925e-02, -1.8911e-03,  2.4073e-02,\n",
      "         2.6313e-02,  2.4564e-02,  2.5070e-03,  2.9270e-02,  2.6210e-02,\n",
      "        -2.9801e-02,  2.6087e-02,  2.1107e-02,  2.1779e-02,  1.2784e-02,\n",
      "         2.3662e-02,  3.3257e-02,  1.3666e-02,  1.4717e-02, -2.5693e-03,\n",
      "         1.4669e-02, -1.9566e-02, -1.7142e-02, -3.0812e-02, -2.1994e-02,\n",
      "         1.7092e-02, -6.8285e-03, -2.3773e-02, -3.2383e-02, -5.0067e-04,\n",
      "         2.5260e-02, -1.5561e-02,  3.2673e-02, -2.9613e-02,  1.6221e-02,\n",
      "         1.6829e-02, -3.6641e-02,  1.8930e-02,  3.1698e-02, -2.5104e-02,\n",
      "        -2.9669e-03, -2.9339e-02, -2.9784e-03,  3.7558e-02, -3.0983e-02,\n",
      "        -5.0861e-03, -5.5774e-03, -1.0976e-02,  2.1974e-02, -1.1801e-02,\n",
      "        -2.1388e-02,  2.2313e-02, -2.7079e-02, -2.7979e-02,  1.6373e-02,\n",
      "        -3.3665e-02, -1.4299e-02, -2.8679e-02, -2.2767e-02, -2.3041e-02,\n",
      "        -2.7506e-02,  3.8096e-03,  4.5947e-03,  2.7333e-02,  9.9457e-04,\n",
      "        -1.3031e-02,  1.1579e-02, -4.3291e-03,  1.4480e-02,  1.8943e-02,\n",
      "        -2.8998e-02, -2.6181e-02, -2.6301e-03,  2.8819e-02,  8.5785e-03,\n",
      "        -5.7729e-03, -2.8097e-02,  1.9599e-02, -6.1758e-03,  1.0292e-02,\n",
      "         2.1661e-02, -2.5944e-02,  2.4791e-03, -8.6287e-03, -8.8444e-03,\n",
      "        -6.0910e-03, -1.7421e-02,  3.5808e-02,  3.5129e-03,  7.5364e-03,\n",
      "        -2.2874e-02,  4.6075e-03,  3.8230e-02, -2.5004e-02,  2.4956e-02,\n",
      "         6.5059e-03, -3.6114e-02,  1.2769e-02,  2.7832e-02, -1.5881e-02,\n",
      "         3.2555e-02,  2.2479e-02, -4.0549e-04, -1.5628e-02, -2.8048e-02,\n",
      "        -8.4990e-03, -2.6972e-02,  2.2830e-03,  3.4891e-03,  2.5916e-02,\n",
      "        -3.0040e-02, -2.5595e-02, -2.8215e-02, -1.6948e-02,  5.8251e-04,\n",
      "         2.8901e-02,  5.8083e-03], device='cuda:0', requires_grad=True)\n"
     ]
    }
   ],
   "source": [
    "layer0_weight = model.get_parameter(\"linear_relu_stack.0.weight\")\n",
    "layer0_bias = model.get_parameter(\"linear_relu_stack.0.bias\")\n",
    "\n",
    "print(f\"layer 0 weight - Shape: {layer0_weight.shape} - {layer0_weight}\\n\")\n",
    "print(f\"layer 0 bias - Shape: {layer0_bias.shape} - {layer0_bias}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Shape of hidden 0: torch.Size([1, 512])\n",
      "Hidden 0: tensor([[ 4.7023e-01,  5.3192e-01,  3.9917e-02,  6.2203e-01,  5.4609e-01,\n",
      "          7.7525e-02,  3.9193e-01,  4.3129e-01,  2.1754e-01, -1.0306e-02,\n",
      "         -9.5101e-02,  3.7093e-01,  2.0274e-01,  9.1209e-01, -3.1995e-01,\n",
      "          1.3447e-01,  2.9983e-01,  3.3304e-01,  1.6213e-01,  1.7942e-01,\n",
      "         -2.0478e-01, -7.1918e-02,  1.7913e-01,  3.6763e-01, -1.7919e-02,\n",
      "          1.1502e+00,  7.4470e-01,  1.0965e-01,  3.4227e-02,  1.4799e-01,\n",
      "          3.1840e-01, -1.1400e-01, -3.8621e-03, -1.6962e-03,  3.1034e-01,\n",
      "         -1.9679e-01,  5.1626e-02, -1.6058e-01,  2.2335e-01,  3.8226e-01,\n",
      "          3.6692e-01, -7.9282e-02, -2.1028e-01, -1.0165e-01, -8.0525e-02,\n",
      "          3.6647e-01, -1.9063e-01, -4.1263e-04,  2.6500e-01,  3.2607e-02,\n",
      "         -1.8676e-01,  3.1039e-01,  5.7008e-01,  6.8967e-01, -5.1019e-02,\n",
      "         -1.4922e-01, -7.0927e-02,  5.9328e-02, -1.9945e-01,  7.5197e-01,\n",
      "          3.4744e-01,  8.5129e-02, -3.0855e-01, -3.8942e-01, -2.0585e-01,\n",
      "         -1.2216e-01, -1.8526e-01,  2.7340e-01,  1.1903e-01,  6.0573e-01,\n",
      "          5.4309e-01,  2.2593e-02,  7.6072e-02, -1.1061e-01, -6.8254e-02,\n",
      "          5.1479e-01, -1.2759e-01, -9.7598e-02,  5.3958e-01, -3.1312e-01,\n",
      "          6.8492e-01,  1.1850e+00,  8.1416e-02, -6.7808e-02,  7.2685e-01,\n",
      "          4.1120e-01,  7.3380e-01,  1.0964e-01,  9.9282e-01, -1.7563e-01,\n",
      "          1.0168e+00,  2.1173e-01,  8.8055e-02,  1.3082e-01,  4.0273e-01,\n",
      "          3.6719e-01,  6.5371e-02,  7.3712e-01, -2.1538e-01, -2.5134e-01,\n",
      "         -6.0957e-02,  2.0184e-01,  2.8441e-01,  4.6177e-01,  1.8682e-02,\n",
      "         -1.4341e-01, -2.0528e-01, -7.9456e-02,  3.6615e-02, -8.4902e-02,\n",
      "          2.8148e-02,  4.8946e-02,  7.8027e-02, -1.9882e-01, -2.4853e-01,\n",
      "          2.5104e-01,  2.4407e-01, -3.5907e-02,  3.7406e-01,  5.2408e-01,\n",
      "         -2.0295e-01,  3.4368e-01,  8.8621e-02,  1.0336e+00, -1.4297e-02,\n",
      "          2.0792e-01, -2.6001e-01,  5.0853e-01, -3.6548e-02, -9.4384e-02,\n",
      "          1.8303e-03, -2.4671e-02,  6.2428e-02,  6.0046e-01, -1.9447e-01,\n",
      "          2.4186e-01, -1.1291e-02,  4.1262e-01, -2.4405e-01, -4.1567e-01,\n",
      "          7.6534e-02,  2.2757e-01,  8.7905e-01,  2.3687e-01,  6.8112e-01,\n",
      "         -1.8854e-02, -2.9410e-01, -1.6427e-01, -1.6262e-02, -2.9834e-01,\n",
      "          3.0460e-01,  5.0293e-01, -1.7275e-01,  4.0566e-02,  3.3185e-01,\n",
      "          8.0558e-02,  2.3555e-03, -1.3054e-01, -1.2488e-01,  1.9208e-01,\n",
      "         -1.7873e-01, -1.4513e-01,  1.9513e-01,  5.7062e-01, -1.0107e-01,\n",
      "         -2.2392e-01, -1.3669e-01, -1.6636e-01,  1.3003e+00, -1.2124e-01,\n",
      "          8.6445e-01,  5.2436e-01, -4.2857e-02,  3.5808e-02, -7.9680e-02,\n",
      "          7.0583e-01,  8.6667e-02,  1.7995e-01,  1.2668e-02, -3.3625e-02,\n",
      "         -1.5186e-01, -2.6788e-01,  1.2059e-01,  4.5703e-01, -3.4576e-02,\n",
      "          3.2075e-01, -6.5648e-02,  3.2002e-01,  1.3570e-01,  1.2484e-01,\n",
      "         -1.1283e-01, -1.3538e-02, -1.5175e-01, -1.4763e-01,  1.1204e-01,\n",
      "          5.8647e-01, -1.2050e-02,  1.0278e+00,  6.6175e-02, -2.0823e-01,\n",
      "         -1.4689e-01,  8.3095e-01, -1.7651e-01,  7.8389e-01,  2.6276e-01,\n",
      "          5.3626e-01, -1.3526e-01, -1.6515e-01, -2.1439e-01,  1.5153e-01,\n",
      "         -1.2853e-01,  1.0641e-02,  4.1958e-01, -3.3195e-02, -1.0868e-01,\n",
      "         -1.5644e-01, -1.6582e-02, -1.9180e-01,  6.4694e-01, -4.3590e-02,\n",
      "         -1.0276e-02,  3.0514e-01, -1.5895e-01, -2.5666e-01, -3.9235e-01,\n",
      "         -2.3501e-01, -2.7715e-01, -9.7797e-02,  2.2450e-01,  3.7300e-01,\n",
      "          6.3365e-01, -2.9643e-01, -5.9127e-02, -2.0551e-01,  9.1445e-01,\n",
      "         -1.4688e-01, -1.6490e-01,  1.4775e-01, -6.6715e-02, -1.7873e-01,\n",
      "         -1.7805e-01,  1.6262e-01, -4.6963e-01, -2.1100e-01,  3.9173e-01,\n",
      "          3.6727e-02,  1.6804e-01,  1.4356e-01,  2.8974e-01,  5.5475e-02,\n",
      "         -1.3614e-01, -2.7336e-02,  4.1034e-01, -2.9730e-02,  1.6796e-01,\n",
      "         -1.2232e-01,  5.2448e-02,  4.7813e-02, -1.3721e-02,  1.1883e-01,\n",
      "         -1.1131e-01, -1.4333e-01,  6.3823e-01,  1.6297e-01,  5.8038e-01,\n",
      "         -1.8669e-02,  5.2228e-01,  5.7841e-01,  2.1165e-01,  3.5641e-01,\n",
      "          3.0308e-01,  1.6612e-01, -7.3250e-02,  8.1580e-02, -1.9019e-01,\n",
      "          1.8634e-01,  7.3987e-02,  2.3609e-01,  6.0848e-02, -6.3229e-02,\n",
      "          1.0723e+00,  6.6015e-02, -7.1423e-02,  4.0757e-01,  6.8669e-01,\n",
      "          7.4917e-01,  1.4156e-01,  4.0179e-01,  6.7158e-04,  2.7358e-01,\n",
      "         -1.3642e-01,  1.1710e-01, -1.5306e-01,  9.2003e-01,  1.2236e-01,\n",
      "         -2.5381e-01,  2.2618e-01,  8.5288e-01,  4.7034e-01,  8.0569e-01,\n",
      "          1.7493e-01, -1.1946e-01, -4.7970e-01, -2.1878e-01, -1.6806e-01,\n",
      "         -2.0478e-01, -2.1865e-01, -3.2975e-01, -6.1302e-02, -1.3524e-02,\n",
      "          1.8854e-01, -7.4840e-02, -3.0316e-01,  1.2380e-01,  9.6901e-02,\n",
      "          7.4512e-01,  5.5836e-02, -2.3004e-01, -7.4771e-02,  3.4344e-01,\n",
      "          5.0808e-02,  1.7684e-01,  1.1832e-01,  2.2489e-01,  4.5490e-01,\n",
      "          1.8628e-02, -1.0541e-01,  3.8785e-01, -2.9897e-01,  1.0112e-01,\n",
      "          4.7267e-01, -1.4987e-02,  6.8437e-01, -1.5393e-01, -1.8595e-02,\n",
      "         -2.0356e-01, -3.8152e-01, -6.9700e-02,  3.4741e-01, -1.4463e-01,\n",
      "          7.9700e-02, -6.3145e-02,  3.9931e-01,  3.5516e-01,  2.6279e-02,\n",
      "          4.7688e-01, -2.2071e-01, -2.2183e-01, -1.6981e-01, -1.0144e-01,\n",
      "         -2.6660e-01, -3.5804e-02, -9.7554e-02,  1.6953e-01,  3.4132e-01,\n",
      "          1.2610e-01, -3.6400e-01, -1.7040e-01,  4.2114e-01,  8.1547e-01,\n",
      "          5.1666e-01,  9.0321e-02, -2.0139e-01, -2.4768e-01,  5.6299e-01,\n",
      "         -1.7795e-01,  5.3265e-01,  7.7726e-01,  5.4966e-01, -8.0128e-02,\n",
      "          6.6438e-01, -8.6292e-02,  4.3962e-01,  5.2623e-01,  3.8460e-01,\n",
      "         -3.3712e-01, -2.7631e-01, -1.6499e-01, -2.5572e-01, -1.3350e-01,\n",
      "         -2.6703e-01,  1.5918e-01,  5.3055e-02, -2.3090e-01, -7.3606e-03,\n",
      "          1.7179e-01, -2.4281e-02,  7.2563e-01, -7.1385e-02,  3.2754e-01,\n",
      "          1.0081e-01,  3.8058e-01,  6.2577e-02,  5.2932e-02,  3.4527e-01,\n",
      "         -1.2625e-01,  4.3203e-02,  2.1212e-01, -2.9975e-01,  3.6120e-01,\n",
      "          8.0152e-01, -1.4690e-01,  2.0745e-01,  2.6826e-01, -1.3455e-01,\n",
      "         -2.6535e-01,  7.3490e-01,  8.9778e-01, -1.1036e-01, -1.3933e-01,\n",
      "         -4.5032e-02,  7.7293e-01,  3.7004e-02, -1.5666e-02,  3.9585e-01,\n",
      "         -1.9762e-01,  3.0592e-01,  4.9439e-01,  3.0675e-01,  2.8767e-01,\n",
      "          8.6526e-01,  1.1771e+00,  4.6883e-01, -1.5155e-01,  1.0614e-01,\n",
      "          4.7378e-01,  8.1141e-02,  2.7832e-01,  1.3031e-01, -9.0395e-02,\n",
      "          8.0560e-02, -1.6769e-01,  2.0827e-01, -1.9631e-01, -1.7689e-01,\n",
      "          4.7223e-01,  2.4582e-01,  1.9485e-01,  7.0188e-02, -3.1070e-01,\n",
      "         -5.6601e-02, -3.2727e-02, -5.4858e-02,  1.2167e-01,  6.7223e-01,\n",
      "         -1.9159e-01, -4.2378e-01, -1.1911e-01, -2.0784e-01, -1.1707e-01,\n",
      "          2.1208e-02, -2.5880e-02,  5.0133e-01, -1.3662e-01, -2.1337e-01,\n",
      "          1.0249e+00,  1.1794e+00, -1.4154e-01,  2.1029e-01,  1.5648e-01,\n",
      "         -9.6374e-02,  6.2023e-01,  3.6251e-01,  2.5602e-01,  6.4640e-01,\n",
      "         -2.9654e-01,  3.4451e-01,  2.5897e-02, -4.8301e-03, -1.4324e-01,\n",
      "         -2.6165e-01,  4.6186e-01,  1.3220e-01,  4.9605e-01, -7.8482e-02,\n",
      "          3.7664e-02,  4.9490e-01,  7.3367e-01, -4.5838e-01,  3.0947e-01,\n",
      "         -3.0509e-01,  9.5611e-01,  2.9430e-01,  5.7978e-01,  1.4336e-02,\n",
      "         -2.1992e-02,  9.3933e-02,  1.6153e-01,  1.3596e-01,  8.0249e-01,\n",
      "          1.8481e-01,  6.1225e-02, -1.3545e-01,  3.7483e-01,  2.4199e-02,\n",
      "          3.7727e-01,  2.8320e-01,  9.9173e-02, -3.5708e-02, -2.3926e-01,\n",
      "          1.2293e-01,  3.6981e-01,  4.3832e-02,  8.6816e-01, -5.5935e-02,\n",
      "         -1.5150e-02,  3.9993e-01,  5.7834e-02, -8.0679e-02, -1.7009e-01,\n",
      "          6.7252e-02, -3.2380e-01]], device='cuda:0', grad_fn=<AddmmBackward0>)\n"
     ]
    }
   ],
   "source": [
    "layer0 = model.linear_relu_stack[0]\n",
    "hidden0 = layer0(flat_image)\n",
    "\n",
    "print(f\"Shape of hidden 0: {hidden0.shape}\")\n",
    "print(f\"Hidden 0: {hidden0}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tensor(0.470225989818573, device='cuda:0', grad_fn=<SelectBackward0>)\n",
      "tensor([0.470225989818573], device='cuda:0', grad_fn=<AddBackward0>)\n"
     ]
    }
   ],
   "source": [
    "torch.set_printoptions(precision=15)\n",
    "print(hidden0[0][0])\n",
    "print(flat_image @ layer0_weight[0] + layer0_bias[0])\n",
    "torch.set_printoptions(precision=4)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "with (\n",
    "  torch.no_grad(),\n",
    "  gzip.GzipFile(\"../models/0-weight.gz\", mode=\"wb\") as f\n",
    "):\n",
    "    f.write(layer0_weight.cpu().numpy().tobytes())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "with (\n",
    "  torch.no_grad(),\n",
    "  gzip.GzipFile(\"../models/0-bias.gz\", mode=\"wb\") as f\n",
    "):\n",
    "    f.write(layer0_bias.cpu().numpy().tobytes())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Shape of hidden 1: torch.Size([1, 512])\n",
      "Hidden 1: tensor([[4.7023e-01, 5.3192e-01, 3.9917e-02, 6.2203e-01, 5.4609e-01, 7.7525e-02,\n",
      "         3.9193e-01, 4.3129e-01, 2.1754e-01, 0.0000e+00, 0.0000e+00, 3.7093e-01,\n",
      "         2.0274e-01, 9.1209e-01, 0.0000e+00, 1.3447e-01, 2.9983e-01, 3.3304e-01,\n",
      "         1.6213e-01, 1.7942e-01, 0.0000e+00, 0.0000e+00, 1.7913e-01, 3.6763e-01,\n",
      "         0.0000e+00, 1.1502e+00, 7.4470e-01, 1.0965e-01, 3.4227e-02, 1.4799e-01,\n",
      "         3.1840e-01, 0.0000e+00, 0.0000e+00, 0.0000e+00, 3.1034e-01, 0.0000e+00,\n",
      "         5.1626e-02, 0.0000e+00, 2.2335e-01, 3.8226e-01, 3.6692e-01, 0.0000e+00,\n",
      "         0.0000e+00, 0.0000e+00, 0.0000e+00, 3.6647e-01, 0.0000e+00, 0.0000e+00,\n",
      "         2.6500e-01, 3.2607e-02, 0.0000e+00, 3.1039e-01, 5.7008e-01, 6.8967e-01,\n",
      "         0.0000e+00, 0.0000e+00, 0.0000e+00, 5.9328e-02, 0.0000e+00, 7.5197e-01,\n",
      "         3.4744e-01, 8.5129e-02, 0.0000e+00, 0.0000e+00, 0.0000e+00, 0.0000e+00,\n",
      "         0.0000e+00, 2.7340e-01, 1.1903e-01, 6.0573e-01, 5.4309e-01, 2.2593e-02,\n",
      "         7.6072e-02, 0.0000e+00, 0.0000e+00, 5.1479e-01, 0.0000e+00, 0.0000e+00,\n",
      "         5.3958e-01, 0.0000e+00, 6.8492e-01, 1.1850e+00, 8.1416e-02, 0.0000e+00,\n",
      "         7.2685e-01, 4.1120e-01, 7.3380e-01, 1.0964e-01, 9.9282e-01, 0.0000e+00,\n",
      "         1.0168e+00, 2.1173e-01, 8.8055e-02, 1.3082e-01, 4.0273e-01, 3.6719e-01,\n",
      "         6.5371e-02, 7.3712e-01, 0.0000e+00, 0.0000e+00, 0.0000e+00, 2.0184e-01,\n",
      "         2.8441e-01, 4.6177e-01, 1.8682e-02, 0.0000e+00, 0.0000e+00, 0.0000e+00,\n",
      "         3.6615e-02, 0.0000e+00, 2.8148e-02, 4.8946e-02, 7.8027e-02, 0.0000e+00,\n",
      "         0.0000e+00, 2.5104e-01, 2.4407e-01, 0.0000e+00, 3.7406e-01, 5.2408e-01,\n",
      "         0.0000e+00, 3.4368e-01, 8.8621e-02, 1.0336e+00, 0.0000e+00, 2.0792e-01,\n",
      "         0.0000e+00, 5.0853e-01, 0.0000e+00, 0.0000e+00, 1.8303e-03, 0.0000e+00,\n",
      "         6.2428e-02, 6.0046e-01, 0.0000e+00, 2.4186e-01, 0.0000e+00, 4.1262e-01,\n",
      "         0.0000e+00, 0.0000e+00, 7.6534e-02, 2.2757e-01, 8.7905e-01, 2.3687e-01,\n",
      "         6.8112e-01, 0.0000e+00, 0.0000e+00, 0.0000e+00, 0.0000e+00, 0.0000e+00,\n",
      "         3.0460e-01, 5.0293e-01, 0.0000e+00, 4.0566e-02, 3.3185e-01, 8.0558e-02,\n",
      "         2.3555e-03, 0.0000e+00, 0.0000e+00, 1.9208e-01, 0.0000e+00, 0.0000e+00,\n",
      "         1.9513e-01, 5.7062e-01, 0.0000e+00, 0.0000e+00, 0.0000e+00, 0.0000e+00,\n",
      "         1.3003e+00, 0.0000e+00, 8.6445e-01, 5.2436e-01, 0.0000e+00, 3.5808e-02,\n",
      "         0.0000e+00, 7.0583e-01, 8.6667e-02, 1.7995e-01, 1.2668e-02, 0.0000e+00,\n",
      "         0.0000e+00, 0.0000e+00, 1.2059e-01, 4.5703e-01, 0.0000e+00, 3.2075e-01,\n",
      "         0.0000e+00, 3.2002e-01, 1.3570e-01, 1.2484e-01, 0.0000e+00, 0.0000e+00,\n",
      "         0.0000e+00, 0.0000e+00, 1.1204e-01, 5.8647e-01, 0.0000e+00, 1.0278e+00,\n",
      "         6.6175e-02, 0.0000e+00, 0.0000e+00, 8.3095e-01, 0.0000e+00, 7.8389e-01,\n",
      "         2.6276e-01, 5.3626e-01, 0.0000e+00, 0.0000e+00, 0.0000e+00, 1.5153e-01,\n",
      "         0.0000e+00, 1.0641e-02, 4.1958e-01, 0.0000e+00, 0.0000e+00, 0.0000e+00,\n",
      "         0.0000e+00, 0.0000e+00, 6.4694e-01, 0.0000e+00, 0.0000e+00, 3.0514e-01,\n",
      "         0.0000e+00, 0.0000e+00, 0.0000e+00, 0.0000e+00, 0.0000e+00, 0.0000e+00,\n",
      "         2.2450e-01, 3.7300e-01, 6.3365e-01, 0.0000e+00, 0.0000e+00, 0.0000e+00,\n",
      "         9.1445e-01, 0.0000e+00, 0.0000e+00, 1.4775e-01, 0.0000e+00, 0.0000e+00,\n",
      "         0.0000e+00, 1.6262e-01, 0.0000e+00, 0.0000e+00, 3.9173e-01, 3.6727e-02,\n",
      "         1.6804e-01, 1.4356e-01, 2.8974e-01, 5.5475e-02, 0.0000e+00, 0.0000e+00,\n",
      "         4.1034e-01, 0.0000e+00, 1.6796e-01, 0.0000e+00, 5.2448e-02, 4.7813e-02,\n",
      "         0.0000e+00, 1.1883e-01, 0.0000e+00, 0.0000e+00, 6.3823e-01, 1.6297e-01,\n",
      "         5.8038e-01, 0.0000e+00, 5.2228e-01, 5.7841e-01, 2.1165e-01, 3.5641e-01,\n",
      "         3.0308e-01, 1.6612e-01, 0.0000e+00, 8.1580e-02, 0.0000e+00, 1.8634e-01,\n",
      "         7.3987e-02, 2.3609e-01, 6.0848e-02, 0.0000e+00, 1.0723e+00, 6.6015e-02,\n",
      "         0.0000e+00, 4.0757e-01, 6.8669e-01, 7.4917e-01, 1.4156e-01, 4.0179e-01,\n",
      "         6.7158e-04, 2.7358e-01, 0.0000e+00, 1.1710e-01, 0.0000e+00, 9.2003e-01,\n",
      "         1.2236e-01, 0.0000e+00, 2.2618e-01, 8.5288e-01, 4.7034e-01, 8.0569e-01,\n",
      "         1.7493e-01, 0.0000e+00, 0.0000e+00, 0.0000e+00, 0.0000e+00, 0.0000e+00,\n",
      "         0.0000e+00, 0.0000e+00, 0.0000e+00, 0.0000e+00, 1.8854e-01, 0.0000e+00,\n",
      "         0.0000e+00, 1.2380e-01, 9.6901e-02, 7.4512e-01, 5.5836e-02, 0.0000e+00,\n",
      "         0.0000e+00, 3.4344e-01, 5.0808e-02, 1.7684e-01, 1.1832e-01, 2.2489e-01,\n",
      "         4.5490e-01, 1.8628e-02, 0.0000e+00, 3.8785e-01, 0.0000e+00, 1.0112e-01,\n",
      "         4.7267e-01, 0.0000e+00, 6.8437e-01, 0.0000e+00, 0.0000e+00, 0.0000e+00,\n",
      "         0.0000e+00, 0.0000e+00, 3.4741e-01, 0.0000e+00, 7.9700e-02, 0.0000e+00,\n",
      "         3.9931e-01, 3.5516e-01, 2.6279e-02, 4.7688e-01, 0.0000e+00, 0.0000e+00,\n",
      "         0.0000e+00, 0.0000e+00, 0.0000e+00, 0.0000e+00, 0.0000e+00, 1.6953e-01,\n",
      "         3.4132e-01, 1.2610e-01, 0.0000e+00, 0.0000e+00, 4.2114e-01, 8.1547e-01,\n",
      "         5.1666e-01, 9.0321e-02, 0.0000e+00, 0.0000e+00, 5.6299e-01, 0.0000e+00,\n",
      "         5.3265e-01, 7.7726e-01, 5.4966e-01, 0.0000e+00, 6.6438e-01, 0.0000e+00,\n",
      "         4.3962e-01, 5.2623e-01, 3.8460e-01, 0.0000e+00, 0.0000e+00, 0.0000e+00,\n",
      "         0.0000e+00, 0.0000e+00, 0.0000e+00, 1.5918e-01, 5.3055e-02, 0.0000e+00,\n",
      "         0.0000e+00, 1.7179e-01, 0.0000e+00, 7.2563e-01, 0.0000e+00, 3.2754e-01,\n",
      "         1.0081e-01, 3.8058e-01, 6.2577e-02, 5.2932e-02, 3.4527e-01, 0.0000e+00,\n",
      "         4.3203e-02, 2.1212e-01, 0.0000e+00, 3.6120e-01, 8.0152e-01, 0.0000e+00,\n",
      "         2.0745e-01, 2.6826e-01, 0.0000e+00, 0.0000e+00, 7.3490e-01, 8.9778e-01,\n",
      "         0.0000e+00, 0.0000e+00, 0.0000e+00, 7.7293e-01, 3.7004e-02, 0.0000e+00,\n",
      "         3.9585e-01, 0.0000e+00, 3.0592e-01, 4.9439e-01, 3.0675e-01, 2.8767e-01,\n",
      "         8.6526e-01, 1.1771e+00, 4.6883e-01, 0.0000e+00, 1.0614e-01, 4.7378e-01,\n",
      "         8.1141e-02, 2.7832e-01, 1.3031e-01, 0.0000e+00, 8.0560e-02, 0.0000e+00,\n",
      "         2.0827e-01, 0.0000e+00, 0.0000e+00, 4.7223e-01, 2.4582e-01, 1.9485e-01,\n",
      "         7.0188e-02, 0.0000e+00, 0.0000e+00, 0.0000e+00, 0.0000e+00, 1.2167e-01,\n",
      "         6.7223e-01, 0.0000e+00, 0.0000e+00, 0.0000e+00, 0.0000e+00, 0.0000e+00,\n",
      "         2.1208e-02, 0.0000e+00, 5.0133e-01, 0.0000e+00, 0.0000e+00, 1.0249e+00,\n",
      "         1.1794e+00, 0.0000e+00, 2.1029e-01, 1.5648e-01, 0.0000e+00, 6.2023e-01,\n",
      "         3.6251e-01, 2.5602e-01, 6.4640e-01, 0.0000e+00, 3.4451e-01, 2.5897e-02,\n",
      "         0.0000e+00, 0.0000e+00, 0.0000e+00, 4.6186e-01, 1.3220e-01, 4.9605e-01,\n",
      "         0.0000e+00, 3.7664e-02, 4.9490e-01, 7.3367e-01, 0.0000e+00, 3.0947e-01,\n",
      "         0.0000e+00, 9.5611e-01, 2.9430e-01, 5.7978e-01, 1.4336e-02, 0.0000e+00,\n",
      "         9.3933e-02, 1.6153e-01, 1.3596e-01, 8.0249e-01, 1.8481e-01, 6.1225e-02,\n",
      "         0.0000e+00, 3.7483e-01, 2.4199e-02, 3.7727e-01, 2.8320e-01, 9.9173e-02,\n",
      "         0.0000e+00, 0.0000e+00, 1.2293e-01, 3.6981e-01, 4.3832e-02, 8.6816e-01,\n",
      "         0.0000e+00, 0.0000e+00, 3.9993e-01, 5.7834e-02, 0.0000e+00, 0.0000e+00,\n",
      "         6.7252e-02, 0.0000e+00]], device='cuda:0', grad_fn=<ReluBackward0>)\n"
     ]
    }
   ],
   "source": [
    "layer1 = model.linear_relu_stack[1]\n",
    "hidden1 = layer1(hidden0)\n",
    "\n",
    "print(f\"Shape of hidden 1: {hidden1.shape}\")\n",
    "print(f\"Hidden 1: {hidden1}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "layer 2 weight - Shape: torch.Size([512, 512]) - Parameter containing:\n",
      "tensor([[ 0.0329,  0.0391,  0.0297,  ..., -0.0288, -0.0236, -0.0275],\n",
      "        [ 0.0211, -0.0174,  0.0346,  ...,  0.0448, -0.0248,  0.0167],\n",
      "        [-0.0371,  0.0114, -0.0073,  ...,  0.0108, -0.0244,  0.0088],\n",
      "        ...,\n",
      "        [-0.0210,  0.0206, -0.0281,  ...,  0.0027,  0.0027,  0.0196],\n",
      "        [-0.0311, -0.0326, -0.0292,  ..., -0.0330,  0.0185,  0.0421],\n",
      "        [-0.0306, -0.0017,  0.0008,  ..., -0.0063,  0.0034, -0.0295]],\n",
      "       device='cuda:0', requires_grad=True)\n",
      "\n",
      "layer 2 bias - Shape: torch.Size([512]) - Parameter containing:\n",
      "tensor([ 9.1274e-03,  5.8128e-02, -1.7652e-02, -2.4247e-02, -2.9891e-02,\n",
      "        -3.3367e-02, -3.5809e-02, -3.3992e-02,  2.4661e-02, -1.9930e-02,\n",
      "        -1.2436e-02,  9.8851e-03,  4.6622e-02,  4.2768e-03, -2.8191e-02,\n",
      "        -2.8528e-02, -2.8024e-02, -3.4333e-02,  1.9791e-02,  4.0003e-02,\n",
      "        -3.5044e-02,  3.3869e-04,  9.7828e-03, -3.1343e-02, -3.6365e-03,\n",
      "        -3.4474e-02, -4.0567e-02,  3.6769e-02,  3.0389e-02, -3.4686e-02,\n",
      "        -1.8927e-02,  3.9603e-02, -1.5784e-02,  4.5054e-02, -5.2551e-03,\n",
      "         3.5398e-02,  5.2619e-03,  1.7942e-02,  4.3591e-02,  2.1473e-04,\n",
      "         4.1001e-02,  1.4748e-02, -3.1869e-02,  1.6008e-02, -2.9309e-02,\n",
      "         2.3134e-02,  1.9477e-02,  4.4587e-02,  1.7112e-02, -1.8141e-02,\n",
      "        -3.8264e-03, -3.0773e-02, -3.5706e-02, -1.9963e-04,  1.3333e-02,\n",
      "         2.7149e-03, -7.8515e-04,  2.6906e-02,  1.3583e-02,  1.9794e-02,\n",
      "        -9.8385e-03, -1.4315e-02,  4.8818e-02,  2.9897e-02,  9.1592e-03,\n",
      "         4.2107e-02, -4.3246e-02, -2.0035e-02, -1.5021e-02, -1.5695e-02,\n",
      "        -2.1387e-02, -1.2522e-03, -2.5309e-02, -1.8791e-02,  2.9949e-02,\n",
      "         7.1430e-03,  5.4464e-03,  3.5723e-02, -2.2096e-02, -2.5737e-02,\n",
      "         3.5545e-02,  1.9317e-02, -3.5250e-02, -2.8756e-02, -3.0737e-02,\n",
      "         3.0240e-05, -1.3054e-02, -9.1521e-03,  4.4519e-02, -2.4287e-02,\n",
      "        -1.6202e-02,  3.2063e-02,  3.9398e-02,  2.8227e-02, -6.5188e-03,\n",
      "         1.0403e-02,  1.7376e-02, -1.9665e-02, -3.3899e-02, -9.2359e-03,\n",
      "         3.1779e-02, -1.8269e-02, -1.6682e-02, -3.6161e-03, -1.1027e-02,\n",
      "        -4.9473e-02, -3.7877e-02, -2.1910e-02,  1.3545e-02,  3.5555e-02,\n",
      "         1.8256e-02, -2.9029e-02,  2.1206e-02,  4.9527e-02,  1.3208e-02,\n",
      "        -4.0549e-02, -1.9918e-02, -3.8366e-02, -1.8188e-02,  2.9223e-02,\n",
      "         1.7430e-02,  2.7322e-02,  1.4887e-02,  2.9485e-02,  3.6525e-02,\n",
      "         2.3157e-02,  4.2897e-02,  2.0677e-02, -2.6272e-02,  2.6511e-02,\n",
      "        -4.1762e-02, -3.7074e-02,  5.7479e-02,  6.4567e-03,  3.8539e-02,\n",
      "         5.3828e-03,  3.2358e-02,  2.4762e-02,  1.5683e-02, -1.8998e-02,\n",
      "        -1.1975e-02,  2.7310e-02, -3.1088e-02,  1.9081e-02, -1.4687e-02,\n",
      "         2.1235e-02,  2.4852e-02,  2.1069e-02,  4.1283e-03,  1.6354e-02,\n",
      "        -3.6577e-02,  6.0827e-02,  2.5559e-02,  4.4401e-02, -3.1918e-02,\n",
      "        -3.5315e-02, -9.5984e-03,  2.7748e-02, -1.8309e-02,  7.8280e-03,\n",
      "         8.8293e-03,  5.1049e-02,  4.1185e-02,  2.7995e-02,  1.2149e-03,\n",
      "        -2.8437e-02, -1.1509e-02, -2.0563e-02,  3.8444e-02, -1.7286e-02,\n",
      "         2.6743e-02,  4.6533e-02,  2.0679e-02, -4.3823e-02,  6.5961e-03,\n",
      "         3.0850e-02,  4.3335e-02, -2.0934e-02, -3.4604e-02, -4.2296e-02,\n",
      "         3.1248e-02, -1.6632e-02,  3.1485e-03, -2.8196e-02, -5.8540e-03,\n",
      "         4.8109e-02,  1.8367e-02,  2.4362e-02, -2.3212e-02,  1.0256e-02,\n",
      "        -4.1174e-02, -9.7288e-03, -8.7374e-03, -2.0021e-02,  2.4372e-03,\n",
      "         1.6332e-02, -2.9728e-02, -2.2221e-02,  2.0266e-02,  1.2572e-02,\n",
      "         6.5338e-03,  3.8196e-02,  1.0151e-02, -3.8347e-03, -2.8974e-02,\n",
      "        -3.5517e-02,  2.1281e-02,  2.7342e-04,  2.0247e-02,  4.4441e-02,\n",
      "        -3.7878e-02, -2.5265e-02, -3.2975e-02,  2.3902e-02,  2.2281e-02,\n",
      "         1.3673e-02,  8.3714e-03, -8.6522e-03, -4.0020e-02, -3.4783e-02,\n",
      "        -1.3030e-02,  5.1062e-02, -3.9442e-02, -1.0479e-02, -2.7386e-02,\n",
      "         4.8437e-02,  1.3345e-02, -6.6865e-03, -1.1711e-02, -2.1520e-02,\n",
      "        -1.1781e-02,  2.0321e-02,  2.9018e-02, -1.6815e-02,  1.3857e-02,\n",
      "        -8.0707e-04,  1.5128e-02,  2.5936e-02,  2.4570e-02,  6.3491e-03,\n",
      "        -3.5936e-02, -2.4742e-02, -2.7323e-02, -1.2155e-02, -4.3767e-02,\n",
      "         3.6752e-02, -3.5192e-02, -2.9449e-02, -2.4886e-02,  1.0387e-02,\n",
      "        -3.6695e-03, -4.5049e-02,  2.5433e-02,  2.1133e-02,  4.7206e-03,\n",
      "         1.7371e-02,  1.9378e-02,  1.3378e-02, -1.0670e-02,  3.0283e-03,\n",
      "         2.2469e-02, -2.8200e-02,  9.2677e-03,  1.3175e-02, -1.7053e-02,\n",
      "        -1.7189e-02,  2.1956e-02, -2.9090e-02, -1.3307e-02, -4.3093e-02,\n",
      "         2.5227e-02, -7.8064e-03, -4.7943e-04,  2.1966e-02,  6.7998e-03,\n",
      "        -3.4129e-02, -4.3925e-02, -1.6492e-02,  2.6755e-02, -1.4569e-03,\n",
      "        -2.7230e-02, -6.3973e-03, -3.2204e-02, -4.3226e-02, -1.9614e-02,\n",
      "         1.3338e-02, -7.7970e-03,  3.1236e-02,  2.4748e-02,  4.7024e-03,\n",
      "         2.8329e-02,  5.2200e-02, -1.6662e-02, -1.4547e-02,  3.4820e-02,\n",
      "        -3.6321e-02,  3.3399e-02,  2.6940e-03, -1.5340e-02, -1.5423e-02,\n",
      "         3.0639e-02, -2.5036e-02,  4.3009e-02, -2.5166e-02, -1.2520e-02,\n",
      "         3.0706e-03,  1.8921e-02,  1.2648e-02,  8.2958e-03,  1.2440e-02,\n",
      "         3.9463e-02, -2.7541e-02, -1.4067e-02,  7.9770e-03, -3.5655e-02,\n",
      "         1.7670e-02, -2.4157e-02,  3.7169e-03,  3.3881e-02, -1.2928e-02,\n",
      "         4.8045e-02,  3.0832e-03,  1.2530e-02, -3.5973e-02, -6.1727e-03,\n",
      "        -4.0460e-02,  4.0003e-02,  2.7560e-02, -3.3154e-02,  2.2384e-02,\n",
      "        -2.1117e-02, -9.1680e-03,  1.1834e-03, -2.6970e-02, -4.5727e-03,\n",
      "         3.6829e-02, -1.7323e-02,  6.4369e-03,  1.2611e-02,  4.5390e-02,\n",
      "         9.5558e-03, -8.0706e-03, -6.9332e-03,  2.2211e-02,  2.2212e-02,\n",
      "        -1.5030e-02, -1.6688e-02,  2.4254e-03,  2.5746e-02,  1.9358e-02,\n",
      "         5.0461e-02, -1.3614e-02,  5.5605e-03, -3.7705e-02, -1.2835e-02,\n",
      "         2.3435e-02,  5.8317e-02, -2.6205e-02,  1.7531e-02,  3.8265e-02,\n",
      "         1.7750e-02, -3.0893e-02,  8.0285e-05,  2.6970e-02, -2.5197e-02,\n",
      "         5.2480e-02, -3.8329e-02,  1.4142e-02, -1.0095e-03, -4.3833e-02,\n",
      "        -1.6500e-02, -1.2842e-02,  1.3397e-02, -5.1757e-03,  2.4811e-02,\n",
      "        -2.7348e-02,  1.0022e-02, -2.6242e-02,  2.7037e-02,  5.3707e-02,\n",
      "         1.6637e-02, -3.3529e-02, -1.5846e-02,  2.4506e-02,  1.6871e-02,\n",
      "         9.9935e-03, -1.6012e-02, -3.2492e-02,  2.8550e-02,  3.3253e-02,\n",
      "        -2.0693e-02, -2.4473e-02,  2.4278e-02, -1.6606e-02,  2.4676e-02,\n",
      "        -3.6970e-02,  1.4752e-02, -4.4621e-02,  4.2851e-02, -2.9027e-03,\n",
      "         3.1315e-02,  3.8485e-02,  4.3992e-03,  4.2078e-02, -2.1720e-03,\n",
      "        -2.3350e-02,  3.4323e-03,  3.9653e-02, -1.6548e-02,  6.8010e-03,\n",
      "        -3.0998e-02, -2.9016e-02,  4.0933e-02,  3.8845e-02,  4.4825e-02,\n",
      "         3.2260e-02,  4.1341e-02, -3.8482e-02,  5.5382e-02,  2.2254e-02,\n",
      "         3.0199e-02,  2.1344e-02,  5.3962e-03,  3.0187e-02, -4.3651e-02,\n",
      "         1.6464e-02, -3.8891e-02,  2.5644e-02,  1.2497e-02,  4.0483e-02,\n",
      "        -4.6026e-02,  3.3178e-02, -2.3052e-02, -1.1453e-02, -2.3346e-02,\n",
      "         1.2618e-02,  6.3210e-03, -1.5277e-03, -2.4449e-02, -2.8400e-02,\n",
      "        -1.8160e-02,  2.1856e-02,  8.7846e-03,  2.2716e-02, -1.8125e-02,\n",
      "        -5.5534e-03,  2.1747e-02, -1.6055e-02, -4.3136e-02,  3.1114e-02,\n",
      "        -2.8810e-02, -1.8497e-02,  5.1485e-02, -3.1917e-02,  1.3562e-02,\n",
      "        -2.2639e-03, -3.6436e-04,  3.1470e-02, -4.0964e-02, -1.1189e-02,\n",
      "        -4.2474e-02,  2.4910e-02, -1.0182e-02,  3.9760e-02,  1.0100e-02,\n",
      "        -1.8937e-02,  3.3047e-04, -1.8044e-02,  3.2018e-02, -3.7700e-02,\n",
      "        -4.1211e-02, -4.0986e-02,  1.2307e-02,  2.7752e-02, -7.4878e-03,\n",
      "         1.1872e-02, -5.6402e-03, -1.6037e-02,  4.0060e-02,  1.7986e-02,\n",
      "        -3.2130e-02, -2.7242e-02, -4.4514e-02, -2.9782e-02, -4.3520e-02,\n",
      "        -2.1882e-02,  5.8299e-02,  3.9287e-02, -2.3766e-02, -4.0741e-02,\n",
      "         1.3666e-03,  2.7649e-02, -3.6321e-02,  4.0957e-02, -1.0558e-02,\n",
      "         3.4046e-02,  1.2063e-02, -2.0452e-02,  1.5892e-02,  4.0104e-02,\n",
      "        -1.1648e-02,  3.3301e-02, -2.7549e-02,  1.4219e-02, -3.5236e-02,\n",
      "        -2.6003e-02, -3.5097e-02,  9.5287e-03, -3.7649e-02, -1.8718e-02,\n",
      "         3.4586e-02,  1.3484e-02], device='cuda:0', requires_grad=True)\n"
     ]
    }
   ],
   "source": [
    "layer2_weight = model.get_parameter(\"linear_relu_stack.2.weight\")\n",
    "layer2_bias = model.get_parameter(\"linear_relu_stack.2.bias\")\n",
    "\n",
    "print(f\"layer 2 weight - Shape: {layer2_weight.shape} - {layer2_weight}\\n\")\n",
    "print(f\"layer 2 bias - Shape: {layer2_bias.shape} - {layer2_bias}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Shape of hidden 2: torch.Size([1, 512])\n",
      "Hidden 2: tensor([[-2.7726e-01, -3.8191e-01,  1.3590e-01, -1.4818e-01, -6.9505e-02,\n",
      "         -8.9749e-02,  5.2888e-01, -1.7615e-01,  1.7339e-01,  2.4589e-01,\n",
      "         -2.2440e-01, -6.3151e-02,  7.5973e-01,  2.8454e-01, -8.5518e-02,\n",
      "          2.3786e-03, -2.5390e-01,  2.2738e-01, -4.0623e-01,  7.1790e-01,\n",
      "         -1.2537e-01, -1.6058e-01, -2.0986e-01,  1.0935e-01,  3.3992e-01,\n",
      "          1.3117e-02, -1.6093e-01, -2.2908e-01, -4.7063e-01, -1.0604e-01,\n",
      "         -1.9042e-01,  9.3610e-01,  7.6163e-01, -1.1777e-01,  7.1609e-02,\n",
      "         -2.0978e-01, -9.0930e-02,  2.8581e-01,  3.8592e-01,  5.4955e-01,\n",
      "         -1.2409e-01,  5.8258e-01,  1.3629e-01,  3.2620e-01,  5.1460e-01,\n",
      "          7.5817e-02, -6.3853e-02,  2.3183e-01,  9.0415e-02,  8.0219e-01,\n",
      "          2.4799e-02,  8.9640e-04, -4.7402e-02, -3.7522e-01,  1.3352e-01,\n",
      "         -1.3034e-01, -3.3467e-02, -5.1212e-03,  3.7537e-01,  1.2864e+00,\n",
      "         -5.8606e-01, -5.3883e-02,  1.7480e-01,  4.0973e-01,  3.7397e-01,\n",
      "         -2.5434e-01,  7.5622e-01,  1.4182e-01, -5.0698e-02,  5.2727e-02,\n",
      "         -1.4849e-02, -4.4554e-01, -1.4451e-01, -2.5104e-01,  2.8993e-01,\n",
      "          1.9059e-01, -1.8939e-01, -1.3510e-01, -3.6893e-01, -7.7749e-02,\n",
      "          1.0012e-01,  5.2910e-01, -5.3479e-02,  5.2904e-01, -1.8591e-01,\n",
      "         -3.5640e-01,  1.0243e-01,  2.1800e-01,  1.4978e+00,  1.5114e+00,\n",
      "         -1.1773e-01, -1.2322e-02,  1.9358e-01, -1.6469e-03,  2.2931e-01,\n",
      "          1.4655e-01,  1.7179e-01,  9.0489e-02,  1.8907e-01, -1.8194e-01,\n",
      "         -1.6471e-01, -6.4641e-02,  2.0003e-01,  4.5678e-01, -6.6888e-02,\n",
      "          3.5150e-02, -2.8849e-02,  6.1036e-01,  6.2388e-02,  1.0125e-02,\n",
      "         -1.3691e-02,  3.9956e-01,  1.7005e-01, -1.2011e-01, -4.0503e-03,\n",
      "         -2.2701e-01, -5.1167e-01,  4.1148e-01, -1.2404e-01,  2.7151e-01,\n",
      "          6.3400e-02,  5.6568e-01,  1.5349e-01,  1.5003e-02,  7.1974e-01,\n",
      "          4.8715e-01, -3.3272e-01,  1.7038e-01, -1.5099e-01,  2.2088e-01,\n",
      "         -2.0083e-02,  2.9783e-02,  9.1641e-01, -2.6314e-01, -2.3118e-01,\n",
      "          5.9749e-01,  1.7519e-01,  6.0396e-02,  3.0483e-01, -3.0893e-01,\n",
      "         -3.6879e-01,  2.9075e-01, -1.5670e-01, -1.3331e-01, -8.4095e-02,\n",
      "         -2.6941e-01, -2.8153e-01,  3.6752e-01,  4.5894e-01,  2.0390e-01,\n",
      "         -1.2717e-01,  6.9603e-01, -1.1581e-02,  1.7892e-01,  7.8689e-01,\n",
      "         -1.1151e-01, -1.9251e-02,  8.3567e-01,  3.8367e-01,  7.4861e-02,\n",
      "          1.6743e-01,  3.4121e-01,  1.3330e-01,  1.7984e-02,  5.6422e-01,\n",
      "         -5.0723e-02, -3.9988e-02, -5.5427e-02,  9.9071e-02, -1.5175e-01,\n",
      "          1.0624e-02,  2.4943e-02,  7.4627e-01,  1.6140e-01, -4.0595e-02,\n",
      "          3.9763e-01, -2.6306e-01, -1.9408e-02,  3.8480e-01,  3.2453e-01,\n",
      "          1.1234e+00, -4.7957e-01,  1.0236e-01,  7.2062e-01, -1.7643e-01,\n",
      "          5.7673e-01,  5.1047e-02, -2.6274e-01,  2.7017e-02, -1.3778e-01,\n",
      "          8.3432e-02,  2.3858e-01,  7.3991e-01,  1.0137e-01, -1.8925e-01,\n",
      "          1.5109e-01, -2.9092e-01, -1.1435e-01, -1.2324e-01, -4.9138e-02,\n",
      "         -5.0185e-01, -5.3760e-02,  3.3206e-01,  1.0545e-01, -1.2771e-01,\n",
      "         -2.9009e-02,  3.4325e-01,  1.4933e-01,  1.2039e-01,  6.3121e-01,\n",
      "         -4.5870e-01,  3.5487e-01, -5.1341e-01,  6.3871e-01,  5.5302e-01,\n",
      "          4.2957e-01, -1.9451e-01,  1.8735e-02, -5.2621e-02,  4.4210e-02,\n",
      "         -1.8929e-01,  4.9247e-01, -1.1043e-01, -1.1315e-01, -5.6996e-02,\n",
      "          1.4376e+00,  4.3946e-01, -1.3213e-02, -1.2859e-01, -3.1412e-01,\n",
      "          2.2783e-01,  1.0571e-01, -4.2715e-02,  2.9212e-01, -2.5307e-01,\n",
      "         -2.9294e-01,  8.5687e-01, -2.8693e-01,  3.3099e-01,  4.8329e-01,\n",
      "          8.3195e-02, -4.0018e-01, -1.1091e-01,  2.5388e-01, -2.4337e-01,\n",
      "          5.1838e-02,  4.5373e-01, -2.5108e-01, -2.5032e-02, -2.4024e-01,\n",
      "         -6.3448e-03, -4.9247e-02,  1.2107e-01, -2.7834e-02, -6.9469e-02,\n",
      "         -2.4515e-01, -7.5022e-02,  4.5087e-01, -2.7296e-02, -7.6290e-03,\n",
      "          3.1286e-01, -2.4186e-01,  3.6272e-01,  1.1864e-01,  2.4954e-01,\n",
      "         -2.9462e-01,  1.2181e-01, -1.5376e-02,  9.8284e-02,  7.6833e-02,\n",
      "          5.4049e-01, -1.7068e-01,  4.7556e-01,  3.7283e-01, -2.9730e-01,\n",
      "          3.5241e-02, -3.8686e-02, -6.5350e-02,  2.1511e-02,  2.1566e-01,\n",
      "          8.2174e-03,  1.9560e-01, -5.7402e-02, -4.9564e-02, -4.2322e-01,\n",
      "         -2.8320e-01,  2.4373e-01, -2.8311e-01, -3.2867e-02,  2.8214e-01,\n",
      "          7.6539e-01,  4.5432e-01,  1.2676e-01, -3.6230e-01,  4.1182e-01,\n",
      "         -3.4550e-01, -1.3160e-01,  1.3221e-01,  3.7971e-02, -2.0089e-02,\n",
      "          1.3072e-01,  1.1578e-01,  7.7795e-02, -1.1400e-01, -2.1911e-01,\n",
      "         -1.1479e-01, -3.5965e-01, -5.7240e-03, -9.8726e-02, -5.6339e-02,\n",
      "          8.6999e-02,  2.8776e-01, -7.2143e-02,  4.7591e-01,  3.4906e-01,\n",
      "          5.6684e-01,  7.2437e-02, -2.0968e-01,  5.9817e-01,  4.8902e-02,\n",
      "          3.3434e-01,  2.9626e-01,  3.8002e-01,  1.5837e-01, -3.0968e-01,\n",
      "         -4.2877e-02, -2.2913e-01,  8.6277e-01,  1.6824e-01, -8.3224e-02,\n",
      "          4.8467e-01,  4.2651e-01,  2.1763e-02,  3.1139e-01,  8.5140e-02,\n",
      "          8.8761e-02,  1.6300e-01,  7.6916e-01,  9.4239e-03,  3.2395e-01,\n",
      "          2.9540e-01,  5.7514e-02, -4.1832e-01,  3.3281e-01,  7.9809e-01,\n",
      "         -2.0955e-01,  2.9336e-02,  2.6024e-01,  6.2136e-01,  3.2972e-01,\n",
      "          7.1315e-02,  8.1047e-01, -4.7684e-01, -6.0710e-02,  1.4040e-02,\n",
      "          8.2228e-02,  1.0372e+00, -2.9849e-01, -3.5030e-01, -1.0731e-01,\n",
      "          1.0769e-02, -2.4940e-01, -2.8667e-01,  2.7471e-01,  2.7012e-01,\n",
      "          4.1103e-01,  1.2152e-01,  3.6849e-01,  1.7208e-01, -1.5510e-01,\n",
      "          2.5054e-01, -9.8700e-02, -1.1289e-01,  1.5201e-01,  5.0065e-01,\n",
      "         -9.5473e-02,  8.7713e-01,  4.9219e-01, -9.6180e-02,  1.0260e+00,\n",
      "         -1.8283e-01, -1.8483e-02, -6.5218e-01,  2.6772e-01, -1.3912e-01,\n",
      "         -5.0621e-02,  3.0937e-01, -4.2671e-01, -3.1573e-01,  5.0007e-02,\n",
      "         -2.1332e-01, -1.3891e-01,  7.1595e-02, -2.9142e-01, -2.9703e-01,\n",
      "          3.5562e-01,  1.8088e-01,  1.0348e-01,  7.7088e-01,  3.0111e-02,\n",
      "          4.6093e-01,  2.3416e-01,  4.5349e-01, -3.9177e-02, -2.2226e-01,\n",
      "          7.6089e-01, -2.0444e-01, -1.9864e-01,  1.4458e-02, -4.2601e-01,\n",
      "          1.3887e-02, -1.7306e-01,  8.3916e-02, -3.2021e-02,  6.1891e-01,\n",
      "         -1.3595e-01,  3.8523e-01, -3.0585e-01,  5.1339e-01, -1.9500e-01,\n",
      "         -1.2351e-01, -1.9131e-03,  7.7510e-02,  1.3769e+00,  5.6448e-02,\n",
      "          2.9588e-01, -1.2591e-01, -1.4258e-01,  7.2140e-01,  8.5881e-01,\n",
      "          1.6014e-01,  4.5824e-01,  7.7585e-01, -1.0446e-01,  1.6473e-01,\n",
      "          2.7142e-01,  1.5699e-01, -7.6091e-03,  8.1916e-02, -2.5153e-01,\n",
      "         -2.9112e-01,  2.8622e-01,  7.3997e-01,  2.3477e-02, -5.1492e-01,\n",
      "          1.0792e-01, -3.7023e-01,  4.4467e-02,  1.4645e-01,  8.0069e-01,\n",
      "          2.2810e-01, -2.3768e-01,  8.8109e-01, -1.3986e-01,  3.5522e-03,\n",
      "          1.6161e-01, -2.4982e-01,  2.1853e-01, -4.6640e-01,  9.0360e-02,\n",
      "         -9.1498e-02,  6.5052e-01,  3.8454e-01,  1.4226e-02,  1.3933e-02,\n",
      "         -4.1434e-01, -2.8242e-03, -2.8966e-01, -1.6385e-01,  1.9585e-01,\n",
      "         -1.2183e-01,  9.1761e-01, -3.8555e-01, -9.4533e-02,  5.7037e-03,\n",
      "         -2.4071e-01, -7.4968e-02, -1.0681e-01,  6.7673e-01,  4.3428e-01,\n",
      "          7.6480e-01, -2.1251e-01,  2.3853e-02,  4.9268e-01, -1.9540e-01,\n",
      "         -3.8314e-01,  2.6935e-01,  5.0334e-01,  2.2128e-01,  4.7452e-01,\n",
      "         -7.8736e-02,  4.0091e-01, -3.9896e-01,  4.0075e-01,  1.7956e-03,\n",
      "          3.9193e-01, -2.9871e-01,  7.5731e-02,  1.6659e-01,  1.6237e-01,\n",
      "          8.8049e-02,  2.8906e-01,  5.2758e-02, -2.0130e-01,  5.6598e-03,\n",
      "         -8.3516e-03,  3.1259e-01,  5.1584e-01, -1.4042e-01,  9.8321e-02,\n",
      "          6.0497e-01, -1.1829e-01]], device='cuda:0', grad_fn=<AddmmBackward0>)\n"
     ]
    }
   ],
   "source": [
    "layer2 = model.linear_relu_stack[2]\n",
    "hidden2 = layer2(hidden1)\n",
    "\n",
    "print(f\"Shape of hidden 2: {hidden2.shape}\")\n",
    "print(f\"Hidden 2: {hidden2}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "with (\n",
    "  torch.no_grad(),\n",
    "  gzip.GzipFile(\"../models/2-weight.gz\", mode=\"wb\") as f\n",
    "):\n",
    "    f.write(layer2_weight.cpu().numpy().tobytes())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "with (\n",
    "  torch.no_grad(),\n",
    "  gzip.GzipFile(\"../models/2-bias.gz\", mode=\"wb\") as f\n",
    "):\n",
    "    f.write(layer2_bias.cpu().numpy().tobytes())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Shape of hidden 3: torch.Size([1, 512])\n",
      "Hidden 3: tensor([[0.0000e+00, 0.0000e+00, 1.3590e-01, 0.0000e+00, 0.0000e+00, 0.0000e+00,\n",
      "         5.2888e-01, 0.0000e+00, 1.7339e-01, 2.4589e-01, 0.0000e+00, 0.0000e+00,\n",
      "         7.5973e-01, 2.8454e-01, 0.0000e+00, 2.3786e-03, 0.0000e+00, 2.2738e-01,\n",
      "         0.0000e+00, 7.1790e-01, 0.0000e+00, 0.0000e+00, 0.0000e+00, 1.0935e-01,\n",
      "         3.3992e-01, 1.3117e-02, 0.0000e+00, 0.0000e+00, 0.0000e+00, 0.0000e+00,\n",
      "         0.0000e+00, 9.3610e-01, 7.6163e-01, 0.0000e+00, 7.1609e-02, 0.0000e+00,\n",
      "         0.0000e+00, 2.8581e-01, 3.8592e-01, 5.4955e-01, 0.0000e+00, 5.8258e-01,\n",
      "         1.3629e-01, 3.2620e-01, 5.1460e-01, 7.5817e-02, 0.0000e+00, 2.3183e-01,\n",
      "         9.0415e-02, 8.0219e-01, 2.4799e-02, 8.9640e-04, 0.0000e+00, 0.0000e+00,\n",
      "         1.3352e-01, 0.0000e+00, 0.0000e+00, 0.0000e+00, 3.7537e-01, 1.2864e+00,\n",
      "         0.0000e+00, 0.0000e+00, 1.7480e-01, 4.0973e-01, 3.7397e-01, 0.0000e+00,\n",
      "         7.5622e-01, 1.4182e-01, 0.0000e+00, 5.2727e-02, 0.0000e+00, 0.0000e+00,\n",
      "         0.0000e+00, 0.0000e+00, 2.8993e-01, 1.9059e-01, 0.0000e+00, 0.0000e+00,\n",
      "         0.0000e+00, 0.0000e+00, 1.0012e-01, 5.2910e-01, 0.0000e+00, 5.2904e-01,\n",
      "         0.0000e+00, 0.0000e+00, 1.0243e-01, 2.1800e-01, 1.4978e+00, 1.5114e+00,\n",
      "         0.0000e+00, 0.0000e+00, 1.9358e-01, 0.0000e+00, 2.2931e-01, 1.4655e-01,\n",
      "         1.7179e-01, 9.0489e-02, 1.8907e-01, 0.0000e+00, 0.0000e+00, 0.0000e+00,\n",
      "         2.0003e-01, 4.5678e-01, 0.0000e+00, 3.5150e-02, 0.0000e+00, 6.1036e-01,\n",
      "         6.2388e-02, 1.0125e-02, 0.0000e+00, 3.9956e-01, 1.7005e-01, 0.0000e+00,\n",
      "         0.0000e+00, 0.0000e+00, 0.0000e+00, 4.1148e-01, 0.0000e+00, 2.7151e-01,\n",
      "         6.3400e-02, 5.6568e-01, 1.5349e-01, 1.5003e-02, 7.1974e-01, 4.8715e-01,\n",
      "         0.0000e+00, 1.7038e-01, 0.0000e+00, 2.2088e-01, 0.0000e+00, 2.9783e-02,\n",
      "         9.1641e-01, 0.0000e+00, 0.0000e+00, 5.9749e-01, 1.7519e-01, 6.0396e-02,\n",
      "         3.0483e-01, 0.0000e+00, 0.0000e+00, 2.9075e-01, 0.0000e+00, 0.0000e+00,\n",
      "         0.0000e+00, 0.0000e+00, 0.0000e+00, 3.6752e-01, 4.5894e-01, 2.0390e-01,\n",
      "         0.0000e+00, 6.9603e-01, 0.0000e+00, 1.7892e-01, 7.8689e-01, 0.0000e+00,\n",
      "         0.0000e+00, 8.3567e-01, 3.8367e-01, 7.4861e-02, 1.6743e-01, 3.4121e-01,\n",
      "         1.3330e-01, 1.7984e-02, 5.6422e-01, 0.0000e+00, 0.0000e+00, 0.0000e+00,\n",
      "         9.9071e-02, 0.0000e+00, 1.0624e-02, 2.4943e-02, 7.4627e-01, 1.6140e-01,\n",
      "         0.0000e+00, 3.9763e-01, 0.0000e+00, 0.0000e+00, 3.8480e-01, 3.2453e-01,\n",
      "         1.1234e+00, 0.0000e+00, 1.0236e-01, 7.2062e-01, 0.0000e+00, 5.7673e-01,\n",
      "         5.1047e-02, 0.0000e+00, 2.7017e-02, 0.0000e+00, 8.3432e-02, 2.3858e-01,\n",
      "         7.3991e-01, 1.0137e-01, 0.0000e+00, 1.5109e-01, 0.0000e+00, 0.0000e+00,\n",
      "         0.0000e+00, 0.0000e+00, 0.0000e+00, 0.0000e+00, 3.3206e-01, 1.0545e-01,\n",
      "         0.0000e+00, 0.0000e+00, 3.4325e-01, 1.4933e-01, 1.2039e-01, 6.3121e-01,\n",
      "         0.0000e+00, 3.5487e-01, 0.0000e+00, 6.3871e-01, 5.5302e-01, 4.2957e-01,\n",
      "         0.0000e+00, 1.8735e-02, 0.0000e+00, 4.4210e-02, 0.0000e+00, 4.9247e-01,\n",
      "         0.0000e+00, 0.0000e+00, 0.0000e+00, 1.4376e+00, 4.3946e-01, 0.0000e+00,\n",
      "         0.0000e+00, 0.0000e+00, 2.2783e-01, 1.0571e-01, 0.0000e+00, 2.9212e-01,\n",
      "         0.0000e+00, 0.0000e+00, 8.5687e-01, 0.0000e+00, 3.3099e-01, 4.8329e-01,\n",
      "         8.3195e-02, 0.0000e+00, 0.0000e+00, 2.5388e-01, 0.0000e+00, 5.1838e-02,\n",
      "         4.5373e-01, 0.0000e+00, 0.0000e+00, 0.0000e+00, 0.0000e+00, 0.0000e+00,\n",
      "         1.2107e-01, 0.0000e+00, 0.0000e+00, 0.0000e+00, 0.0000e+00, 4.5087e-01,\n",
      "         0.0000e+00, 0.0000e+00, 3.1286e-01, 0.0000e+00, 3.6272e-01, 1.1864e-01,\n",
      "         2.4954e-01, 0.0000e+00, 1.2181e-01, 0.0000e+00, 9.8284e-02, 7.6833e-02,\n",
      "         5.4049e-01, 0.0000e+00, 4.7556e-01, 3.7283e-01, 0.0000e+00, 3.5241e-02,\n",
      "         0.0000e+00, 0.0000e+00, 2.1511e-02, 2.1566e-01, 8.2174e-03, 1.9560e-01,\n",
      "         0.0000e+00, 0.0000e+00, 0.0000e+00, 0.0000e+00, 2.4373e-01, 0.0000e+00,\n",
      "         0.0000e+00, 2.8214e-01, 7.6539e-01, 4.5432e-01, 1.2676e-01, 0.0000e+00,\n",
      "         4.1182e-01, 0.0000e+00, 0.0000e+00, 1.3221e-01, 3.7971e-02, 0.0000e+00,\n",
      "         1.3072e-01, 1.1578e-01, 7.7795e-02, 0.0000e+00, 0.0000e+00, 0.0000e+00,\n",
      "         0.0000e+00, 0.0000e+00, 0.0000e+00, 0.0000e+00, 8.6999e-02, 2.8776e-01,\n",
      "         0.0000e+00, 4.7591e-01, 3.4906e-01, 5.6684e-01, 7.2437e-02, 0.0000e+00,\n",
      "         5.9817e-01, 4.8902e-02, 3.3434e-01, 2.9626e-01, 3.8002e-01, 1.5837e-01,\n",
      "         0.0000e+00, 0.0000e+00, 0.0000e+00, 8.6277e-01, 1.6824e-01, 0.0000e+00,\n",
      "         4.8467e-01, 4.2651e-01, 2.1763e-02, 3.1139e-01, 8.5140e-02, 8.8761e-02,\n",
      "         1.6300e-01, 7.6916e-01, 9.4239e-03, 3.2395e-01, 2.9540e-01, 5.7514e-02,\n",
      "         0.0000e+00, 3.3281e-01, 7.9809e-01, 0.0000e+00, 2.9336e-02, 2.6024e-01,\n",
      "         6.2136e-01, 3.2972e-01, 7.1315e-02, 8.1047e-01, 0.0000e+00, 0.0000e+00,\n",
      "         1.4040e-02, 8.2228e-02, 1.0372e+00, 0.0000e+00, 0.0000e+00, 0.0000e+00,\n",
      "         1.0769e-02, 0.0000e+00, 0.0000e+00, 2.7471e-01, 2.7012e-01, 4.1103e-01,\n",
      "         1.2152e-01, 3.6849e-01, 1.7208e-01, 0.0000e+00, 2.5054e-01, 0.0000e+00,\n",
      "         0.0000e+00, 1.5201e-01, 5.0065e-01, 0.0000e+00, 8.7713e-01, 4.9219e-01,\n",
      "         0.0000e+00, 1.0260e+00, 0.0000e+00, 0.0000e+00, 0.0000e+00, 2.6772e-01,\n",
      "         0.0000e+00, 0.0000e+00, 3.0937e-01, 0.0000e+00, 0.0000e+00, 5.0007e-02,\n",
      "         0.0000e+00, 0.0000e+00, 7.1595e-02, 0.0000e+00, 0.0000e+00, 3.5562e-01,\n",
      "         1.8088e-01, 1.0348e-01, 7.7088e-01, 3.0111e-02, 4.6093e-01, 2.3416e-01,\n",
      "         4.5349e-01, 0.0000e+00, 0.0000e+00, 7.6089e-01, 0.0000e+00, 0.0000e+00,\n",
      "         1.4458e-02, 0.0000e+00, 1.3887e-02, 0.0000e+00, 8.3916e-02, 0.0000e+00,\n",
      "         6.1891e-01, 0.0000e+00, 3.8523e-01, 0.0000e+00, 5.1339e-01, 0.0000e+00,\n",
      "         0.0000e+00, 0.0000e+00, 7.7510e-02, 1.3769e+00, 5.6448e-02, 2.9588e-01,\n",
      "         0.0000e+00, 0.0000e+00, 7.2140e-01, 8.5881e-01, 1.6014e-01, 4.5824e-01,\n",
      "         7.7585e-01, 0.0000e+00, 1.6473e-01, 2.7142e-01, 1.5699e-01, 0.0000e+00,\n",
      "         8.1916e-02, 0.0000e+00, 0.0000e+00, 2.8622e-01, 7.3997e-01, 2.3477e-02,\n",
      "         0.0000e+00, 1.0792e-01, 0.0000e+00, 4.4467e-02, 1.4645e-01, 8.0069e-01,\n",
      "         2.2810e-01, 0.0000e+00, 8.8109e-01, 0.0000e+00, 3.5522e-03, 1.6161e-01,\n",
      "         0.0000e+00, 2.1853e-01, 0.0000e+00, 9.0360e-02, 0.0000e+00, 6.5052e-01,\n",
      "         3.8454e-01, 1.4226e-02, 1.3933e-02, 0.0000e+00, 0.0000e+00, 0.0000e+00,\n",
      "         0.0000e+00, 1.9585e-01, 0.0000e+00, 9.1761e-01, 0.0000e+00, 0.0000e+00,\n",
      "         5.7037e-03, 0.0000e+00, 0.0000e+00, 0.0000e+00, 6.7673e-01, 4.3428e-01,\n",
      "         7.6480e-01, 0.0000e+00, 2.3853e-02, 4.9268e-01, 0.0000e+00, 0.0000e+00,\n",
      "         2.6935e-01, 5.0334e-01, 2.2128e-01, 4.7452e-01, 0.0000e+00, 4.0091e-01,\n",
      "         0.0000e+00, 4.0075e-01, 1.7956e-03, 3.9193e-01, 0.0000e+00, 7.5731e-02,\n",
      "         1.6659e-01, 1.6237e-01, 8.8049e-02, 2.8906e-01, 5.2758e-02, 0.0000e+00,\n",
      "         5.6598e-03, 0.0000e+00, 3.1259e-01, 5.1584e-01, 0.0000e+00, 9.8321e-02,\n",
      "         6.0497e-01, 0.0000e+00]], device='cuda:0', grad_fn=<ReluBackward0>)\n"
     ]
    }
   ],
   "source": [
    "layer3 = model.linear_relu_stack[3]\n",
    "hidden3 = layer3(hidden2)\n",
    "\n",
    "print(f\"Shape of hidden 3: {hidden3.shape}\")\n",
    "print(f\"Hidden 3: {hidden3}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "layer 4 weight - Shape: torch.Size([10, 512]) - Parameter containing:\n",
      "tensor([[-0.0078,  0.0882,  0.0149,  ..., -0.0204, -0.0397,  0.0218],\n",
      "        [ 0.0101,  0.1217, -0.0114,  ..., -0.0185,  0.0421,  0.0109],\n",
      "        [-0.0172, -0.0295,  0.0146,  ...,  0.0151, -0.0309, -0.0137],\n",
      "        ...,\n",
      "        [ 0.0016, -0.0628,  0.0303,  ...,  0.0086,  0.0134,  0.0081],\n",
      "        [ 0.0226, -0.0875,  0.0129,  ..., -0.0326,  0.0187,  0.0264],\n",
      "        [ 0.0371, -0.0418,  0.0161,  ..., -0.0260,  0.0421, -0.0426]],\n",
      "       device='cuda:0', requires_grad=True)\n",
      "\n",
      "layer 4 bias - Shape: torch.Size([10]) - Parameter containing:\n",
      "tensor([-0.0409, -0.0246, -0.0050,  0.0328, -0.0801,  0.1937,  0.0077,  0.0859,\n",
      "        -0.0652, -0.0485], device='cuda:0', requires_grad=True)\n"
     ]
    }
   ],
   "source": [
    "layer4_weight = model.get_parameter(\"linear_relu_stack.4.weight\")\n",
    "layer4_bias = model.get_parameter(\"linear_relu_stack.4.bias\")\n",
    "\n",
    "print(f\"layer 4 weight - Shape: {layer4_weight.shape} - {layer4_weight}\\n\")\n",
    "print(f\"layer 4 bias - Shape: {layer4_bias.shape} - {layer4_bias}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Shape of hidden 4: torch.Size([1, 10])\n",
      "Hidden 4: tensor([[-2.2995, -2.1135, -1.0424, -1.6524, -0.9015,  2.4404, -1.1452,  2.6658,\n",
      "          1.6956,  3.0893]], device='cuda:0', grad_fn=<AddmmBackward0>)\n"
     ]
    }
   ],
   "source": [
    "layer4 = model.linear_relu_stack[4]\n",
    "hidden4 = layer4(hidden3)\n",
    "\n",
    "print(f\"Shape of hidden 4: {hidden4.shape}\")\n",
    "print(f\"Hidden 4: {hidden4}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "with (\n",
    "  torch.no_grad(),\n",
    "  gzip.GzipFile(\"../models/4-weight.gz\", mode=\"wb\") as f\n",
    "):\n",
    "    f.write(layer4_weight.cpu().numpy().tobytes())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "with (\n",
    "  torch.no_grad(),\n",
    "  gzip.GzipFile(\"../models/4-bias.gz\", mode=\"wb\") as f\n",
    "):\n",
    "    f.write(layer4_bias.cpu().numpy().tobytes())"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "![](webgl-memory.png)"
   ]
  }
 ],
 "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": "3.11.3"
  },
  "orig_nbformat": 4
 },
 "nbformat": 4,
 "nbformat_minor": 2
}