Refactor tutorial code

Author: KazunoriSumiya

tutorials/notebooks/mct_features_notebooks/pytorch/example_pytorch_XQuant_Extension_Tool.ipynb

@@ -18,18 +18,18 @@
     "## Summary\n",
     "We will cover the following steps:\n",
     "\n",
-    "1. Load a pre-trained MobileNetV3 model\n",
-    "2. Perform Post-Training Quantization using MCT (default parameter)\n",
-    "3. General Troubleshooting\n",
+    "1. MobileNet V3 Setting\n",
+    "2. Dataset Preparation\n",
+    "3. Perform Post-Training Quantization using MCT (default parameter)\n",
+    "4. General Troubleshooting\n",
     "   - Representative Dataset Size & Diversity\n",
     "   - Bias Correction\n",
     "   - Using More Samples in Mixed Precision Quantization\n",
     "   - Threshold Selection Error Method\n",
     "   - Enabling Hessian-Based Mixed Precision\n",
     "   - GPTQ - Gradient-Based Post-Training Quantization\n",
-    "4. Conclusion\n",
+    "5. Conclusion\n",
     "\n",
-    "    \n",
     "## Setup\n",
     "Install the relevant packages:"
    ]
@@ -99,9 +99,13 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "## Representative Dataset\n",
-    "Download ImageNet dataset.\n",
-    "This step may take several minutes..."
+    "## Dataset Preparation\n",
+    "### Download ImageNet validation set\n",
+    "Download ImageNet dataset (validation split only).\n",
+    "\n",
+    "This step may take several minutes...\n",
+    "\n",
+    "**Note:** For demonstration purposes, we use the validation set for the model quantization routines. Usually, a subset of the training dataset is used, but loading it is a heavy procedure that is unnecessary for the sake of this demonstration."
    ]
   },
   {
@@ -113,14 +117,48 @@
     "if not os.path.isdir('imagenet'):\n",
     "    !mkdir imagenet\n",
     "    !wget -P imagenet https://image-net.org/data/ILSVRC/2012/ILSVRC2012_devkit_t12.tar.gz\n",
-    "    !wget -P imagenet https://image-net.org/data/ILSVRC/2012/ILSVRC2012_img_val.tar\n",
-    "\n",
-    "dataset_path = './imagenet'\n",
-    "dataset = ImageNet(root=dataset_path, split='val', transform=weights.transforms())\n",
+    "    !wget -P imagenet https://image-net.org/data/ILSVRC/2012/ILSVRC2012_img_val.tar"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Extract ImageNet validation dataset using torchvision \"datasets\" module."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "dataset = ImageNet(root='./imagenet', split='val', transform=weights.transforms())"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Representative Dataset\n",
+    "For quantization with MCT, we need to define a representative dataset. This dataset is a generator that returns a list of images:"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
     "batch_size = 16\n",
     "n_iter = 10\n",
     "\n",
-    "dataloader = DataLoader(dataset, batch_size=batch_size, shuffle=True)"
+    "dataloader = DataLoader(dataset, batch_size=batch_size, shuffle=True)\n",
+    "\n",
+    "def representative_dataset_gen():\n",
+    "    dataloader_iter = iter(dataloader)\n",
+    "    for _ in range(n_iter):\n",
+    "        yield [next(dataloader_iter)[0]]"
    ]
   },
   {
@@ -151,18 +189,6 @@
     "    return dataloader"
    ]
   },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "def representative_dataset_gen():\n",
-    "    dataloader_iter = iter(dataloader)\n",
-    "    for _ in range(n_iter):\n",
-    "        yield [next(dataloader_iter)[0]]"
-   ]
-  },
   {
    "cell_type": "markdown",
    "metadata": {},
@@ -547,8 +573,7 @@
     "    results.append(val_acc)\n",
     "    torch.cuda.empty_cache()\n",
     "    del quantized_model\n",
-    "    gc.collect()\n",
-    "    "
+    "    gc.collect()    "
    ]
   },
   {
@@ -619,7 +644,7 @@
    "metadata": {},
    "source": [
     "## Conclusion\n",
-    "These analyses showed that accuracy improved by 0.31% when the number of images was 80, and by 0.54% when the number of GPTQ epochs was 80, resulting in a reduced quantization accuracy loss.By following these troubleshooting steps, you can improve the accuracy of your quantized model."
+    "These analyses showed that accuracy improved by 0.31% when the number of images was 80, and by 0.54% when the number of GPTQ epochs was 80, resulting in a reduced quantization accuracy loss. By following these troubleshooting steps can help improve the accuracy of your quantized model."
    ]
   },
   {