{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "from PIL import Image\n",
    "import pandas as pd\n",
    "import os\n",
    "from datasets import Dataset, Image, DatasetDict\n",
    "from torchvision.transforms import RandomResizedCrop, Compose, Normalize, ToTensor\n",
    "from transformers import (\n",
    "    AutoImageProcessor,\n",
    "    AutoModelForImageClassification,\n",
    "    TrainingArguments,\n",
    "    Trainer,\n",
    "    DefaultDataCollator,\n",
    ")\n",
    "import evaluate\n",
    "import numpy as np"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Load data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "file2obj = pd.read_csv(\"../data/processed/OM_file_to_obj.csv\")\n",
    "file2obj[\"image\"] = file2obj.apply(lambda x: os.path.join(\"..\", x[\"root\"], x[\"file\"]), axis=1)\n",
    "file2obj.rename(columns={\"obj_num\": \"label\"}, inplace=True)\n",
    "\n",
    "# Group by 'obj_num' and count occurrences\n",
    "obj_num_counts = file2obj[\"label\"].value_counts()\n",
    "\n",
    "# Filter rows where 'obj_num' appears more than twice\n",
    "file2obj_3 = file2obj[file2obj[\"label\"].isin(obj_num_counts[obj_num_counts > 2].index)]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Form HF dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "59370086a1b64dc5842d9becd9019aad",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Casting to class labels:   0%|          | 0/25725 [00:00<?, ? examples/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "ds = Dataset.from_pandas(file2obj_3[[\"image\", \"label\"]], preserve_index=False).cast_column(\n",
    "    \"image\", Image()\n",
    ")\n",
    "ds = ds.class_encode_column(\"label\")\n",
    "trainval_test = ds.train_test_split(stratify_by_column=\"label\", test_size=0.16)\n",
    "train_val = trainval_test[\"train\"].train_test_split(stratify_by_column=\"label\", test_size=16 / 84)\n",
    "ds = DatasetDict(\n",
    "    {\"train\": train_val[\"train\"], \"valid\": train_val[\"test\"], \"test\": trainval_test[\"test\"]}\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Transform data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "checkpoint = \"google/efficientnet-b3\"\n",
    "image_processor = AutoImageProcessor.from_pretrained(checkpoint)\n",
    "\n",
    "\n",
    "normalize = Normalize(mean=image_processor.image_mean, std=image_processor.image_std)\n",
    "size = (\n",
    "    image_processor.size[\"shortest_edge\"]\n",
    "    if \"shortest_edge\" in image_processor.size\n",
    "    else (image_processor.size[\"height\"], image_processor.size[\"width\"])\n",
    ")\n",
    "_transforms = Compose([RandomResizedCrop(size), ToTensor(), normalize])\n",
    "\n",
    "\n",
    "def transforms(examples):\n",
    "    examples[\"pixel_values\"] = [_transforms(img.convert(\"RGB\")) for img in examples[\"image\"]]\n",
    "    del examples[\"image\"]\n",
    "    return examples\n",
    "\n",
    "\n",
    "ds = ds.with_transform(transforms)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Set up model and metrics"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Some weights of EfficientNetForImageClassification were not initialized from the model checkpoint at google/efficientnet-b3 and are newly initialized because the shapes did not match:\n",
      "- classifier.weight: found shape torch.Size([1000, 1536]) in the checkpoint and torch.Size([3872, 1536]) in the model instantiated\n",
      "- classifier.bias: found shape torch.Size([1000]) in the checkpoint and torch.Size([3872]) in the model instantiated\n",
      "You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.\n"
     ]
    }
   ],
   "source": [
    "labels = ds[\"train\"].features[\"label\"].names\n",
    "model = AutoModelForImageClassification.from_pretrained(\n",
    "    checkpoint,\n",
    "    num_labels=len(labels),\n",
    "    id2label={str(i): c for i, c in enumerate(labels)},\n",
    "    label2id={c: str(i) for i, c in enumerate(labels)},\n",
    "    ignore_mismatched_sizes=True,\n",
    ")\n",
    "\n",
    "data_collator = DefaultDataCollator()\n",
    "\n",
    "accuracy = evaluate.load(\"accuracy\")\n",
    "\n",
    "\n",
    "def compute_metrics(eval_pred):\n",
    "    predictions, labels = eval_pred\n",
    "    predictions = np.argmax(predictions, axis=1)\n",
    "    return accuracy.compute(predictions=predictions, references=labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Train model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\u001b[34m\u001b[1mwandb\u001b[0m: \u001b[33mWARNING\u001b[0m Serializing object of type dict that is 147552 bytes\n",
      "\u001b[34m\u001b[1mwandb\u001b[0m: \u001b[33mWARNING\u001b[0m Serializing object of type dict that is 147552 bytes\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "cb34337db5584dfbbc4a76bb7e724b26",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/274 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'loss': 8.0521, 'learning_rate': 1.785714285714286e-05, 'epoch': 0.04}\n",
      "{'loss': 8.0927, 'learning_rate': 3.571428571428572e-05, 'epoch': 0.07}\n",
      "{'loss': 8.1187, 'learning_rate': 4.959349593495935e-05, 'epoch': 0.11}\n",
      "{'loss': 8.2335, 'learning_rate': 4.75609756097561e-05, 'epoch': 0.15}\n",
      "{'loss': 8.2531, 'learning_rate': 4.5528455284552844e-05, 'epoch': 0.18}\n",
      "{'loss': 8.2873, 'learning_rate': 4.3495934959349595e-05, 'epoch': 0.22}\n",
      "{'loss': 8.2071, 'learning_rate': 4.146341463414634e-05, 'epoch': 0.26}\n",
      "{'loss': 8.2287, 'learning_rate': 3.943089430894309e-05, 'epoch': 0.29}\n",
      "{'loss': 8.1928, 'learning_rate': 3.739837398373984e-05, 'epoch': 0.33}\n",
      "{'loss': 8.2053, 'learning_rate': 3.5365853658536584e-05, 'epoch': 0.36}\n",
      "{'loss': 8.1621, 'learning_rate': 3.3333333333333335e-05, 'epoch': 0.4}\n",
      "{'loss': 8.1731, 'learning_rate': 3.130081300813008e-05, 'epoch': 0.44}\n",
      "{'loss': 8.1447, 'learning_rate': 2.926829268292683e-05, 'epoch': 0.47}\n",
      "{'loss': 8.1161, 'learning_rate': 2.7235772357723577e-05, 'epoch': 0.51}\n",
      "{'loss': 8.1081, 'learning_rate': 2.5203252032520324e-05, 'epoch': 0.55}\n",
      "{'loss': 8.0828, 'learning_rate': 2.3170731707317075e-05, 'epoch': 0.58}\n",
      "{'loss': 8.1312, 'learning_rate': 2.1138211382113822e-05, 'epoch': 0.62}\n",
      "{'loss': 8.1627, 'learning_rate': 1.9105691056910573e-05, 'epoch': 0.66}\n",
      "{'loss': 8.0692, 'learning_rate': 1.707317073170732e-05, 'epoch': 0.69}\n",
      "{'loss': 7.9943, 'learning_rate': 1.5040650406504067e-05, 'epoch': 0.73}\n",
      "{'loss': 8.0139, 'learning_rate': 1.3008130081300815e-05, 'epoch': 0.77}\n",
      "{'loss': 8.027, 'learning_rate': 1.0975609756097562e-05, 'epoch': 0.8}\n",
      "{'loss': 8.0628, 'learning_rate': 8.94308943089431e-06, 'epoch': 0.84}\n",
      "{'loss': 8.1264, 'learning_rate': 6.910569105691057e-06, 'epoch': 0.88}\n",
      "{'loss': 8.0408, 'learning_rate': 4.8780487804878055e-06, 'epoch': 0.91}\n",
      "{'loss': 8.071, 'learning_rate': 2.8455284552845528e-06, 'epoch': 0.95}\n",
      "{'loss': 8.1237, 'learning_rate': 8.130081300813009e-07, 'epoch': 0.99}\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "5915cfa41d474a399ce0f53bc8f6f947",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/65 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'eval_loss': 8.02699089050293, 'eval_accuracy': 0.02575315840621963, 'eval_runtime': 25.2001, 'eval_samples_per_second': 163.333, 'eval_steps_per_second': 2.579, 'epoch': 1.0}\n",
      "{'train_runtime': 236.2359, 'train_samples_per_second': 74.049, 'train_steps_per_second': 1.16, 'train_loss': 8.129460439194728, 'epoch': 1.0}\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "TrainOutput(global_step=274, training_loss=8.129460439194728, metrics={'train_runtime': 236.2359, 'train_samples_per_second': 74.049, 'train_steps_per_second': 1.16, 'train_loss': 8.129460439194728, 'epoch': 1.0})"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "training_args = TrainingArguments(\n",
    "    output_dir=\"../models/test\",\n",
    "    remove_unused_columns=False,\n",
    "    evaluation_strategy=\"epoch\",\n",
    "    save_strategy=\"epoch\",\n",
    "    learning_rate=5e-5,\n",
    "    per_device_train_batch_size=64,\n",
    "    # gradient_accumulation_steps=2,\n",
    "    per_device_eval_batch_size=64,\n",
    "    num_train_epochs=1,\n",
    "    warmup_ratio=0.1,\n",
    "    logging_steps=10,\n",
    "    load_best_model_at_end=True,\n",
    "    metric_for_best_model=\"accuracy\",\n",
    "    push_to_hub=False,\n",
    "\n",
    ")\n",
    "\n",
    "trainer = Trainer(\n",
    "    model=model,\n",
    "    args=training_args,\n",
    "    train_dataset=ds[\"train\"],  # .select(range(100)),\n",
    "    eval_dataset=ds[\"valid\"],  # .select(range(100)),\n",
    "    tokenizer=image_processor,\n",
    "    compute_metrics=compute_metrics,\n",
    "    data_collator=data_collator,\n",
    ")\n",
    "\n",
    "trainer.train()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Evaluation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "4979f1d5536f4a3e97ecbc36c7eebbfa",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/7 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'eval_loss': 8.275933265686035, 'eval_accuracy': 0.0, 'eval_runtime': 0.6419, 'eval_samples_per_second': 155.791, 'eval_steps_per_second': 10.905, 'epoch': 0.57}\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "cf6f48e995bf427db3c86d1d988bf752",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/7 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "results = trainer.evaluate()\n",
    "print(results)\n",
    "\n",
    "test_results = trainer.predict(ds[\"test\"].select(range(100)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "EfficientNetForImageClassification(\n",
       "  (efficientnet): EfficientNetModel(\n",
       "    (embeddings): EfficientNetEmbeddings(\n",
       "      (padding): ZeroPad2d((0, 1, 0, 1))\n",
       "      (convolution): Conv2d(3, 40, kernel_size=(3, 3), stride=(2, 2), padding=valid, bias=False)\n",
       "      (batchnorm): BatchNorm2d(40, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "      (activation): SiLU()\n",
       "    )\n",
       "    (encoder): EfficientNetEncoder(\n",
       "      (blocks): ModuleList(\n",
       "        (0): EfficientNetBlock(\n",
       "          (depthwise_conv): EfficientNetDepthwiseLayer(\n",
       "            (depthwise_conv_pad): ZeroPad2d((0, 1, 0, 1))\n",
       "            (depthwise_conv): EfficientNetDepthwiseConv2d(40, 40, kernel_size=(3, 3), stride=(1, 1), padding=same, groups=40, bias=False)\n",
       "            (depthwise_norm): BatchNorm2d(40, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (depthwise_act): SiLU()\n",
       "          )\n",
       "          (squeeze_excite): EfficientNetSqueezeExciteLayer(\n",
       "            (squeeze): AdaptiveAvgPool2d(output_size=1)\n",
       "            (reduce): Conv2d(40, 10, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (expand): Conv2d(10, 40, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (act_reduce): SiLU()\n",
       "            (act_expand): Sigmoid()\n",
       "          )\n",
       "          (projection): EfficientNetFinalBlockLayer(\n",
       "            (project_conv): Conv2d(40, 24, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (project_bn): BatchNorm2d(24, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (dropout): Dropout(p=0.0, inplace=False)\n",
       "          )\n",
       "        )\n",
       "        (1): EfficientNetBlock(\n",
       "          (depthwise_conv): EfficientNetDepthwiseLayer(\n",
       "            (depthwise_conv_pad): ZeroPad2d((0, 1, 0, 1))\n",
       "            (depthwise_conv): EfficientNetDepthwiseConv2d(24, 24, kernel_size=(3, 3), stride=(1, 1), padding=same, groups=24, bias=False)\n",
       "            (depthwise_norm): BatchNorm2d(24, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (depthwise_act): SiLU()\n",
       "          )\n",
       "          (squeeze_excite): EfficientNetSqueezeExciteLayer(\n",
       "            (squeeze): AdaptiveAvgPool2d(output_size=1)\n",
       "            (reduce): Conv2d(24, 6, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (expand): Conv2d(6, 24, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (act_reduce): SiLU()\n",
       "            (act_expand): Sigmoid()\n",
       "          )\n",
       "          (projection): EfficientNetFinalBlockLayer(\n",
       "            (project_conv): Conv2d(24, 24, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (project_bn): BatchNorm2d(24, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (dropout): Dropout(p=0.007692307692307693, inplace=False)\n",
       "          )\n",
       "        )\n",
       "        (2): EfficientNetBlock(\n",
       "          (expansion): EfficientNetExpansionLayer(\n",
       "            (expand_conv): Conv2d(24, 144, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (expand_bn): BatchNorm2d(144, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)\n",
       "            (expand_act): SiLU()\n",
       "          )\n",
       "          (depthwise_conv): EfficientNetDepthwiseLayer(\n",
       "            (depthwise_conv_pad): ZeroPad2d((0, 1, 0, 1))\n",
       "            (depthwise_conv): EfficientNetDepthwiseConv2d(144, 144, kernel_size=(3, 3), stride=(2, 2), padding=valid, groups=144, bias=False)\n",
       "            (depthwise_norm): BatchNorm2d(144, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (depthwise_act): SiLU()\n",
       "          )\n",
       "          (squeeze_excite): EfficientNetSqueezeExciteLayer(\n",
       "            (squeeze): AdaptiveAvgPool2d(output_size=1)\n",
       "            (reduce): Conv2d(144, 6, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (expand): Conv2d(6, 144, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (act_reduce): SiLU()\n",
       "            (act_expand): Sigmoid()\n",
       "          )\n",
       "          (projection): EfficientNetFinalBlockLayer(\n",
       "            (project_conv): Conv2d(144, 32, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (project_bn): BatchNorm2d(32, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (dropout): Dropout(p=0.015384615384615385, inplace=False)\n",
       "          )\n",
       "        )\n",
       "        (3): EfficientNetBlock(\n",
       "          (expansion): EfficientNetExpansionLayer(\n",
       "            (expand_conv): Conv2d(32, 192, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (expand_bn): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)\n",
       "            (expand_act): SiLU()\n",
       "          )\n",
       "          (depthwise_conv): EfficientNetDepthwiseLayer(\n",
       "            (depthwise_conv_pad): ZeroPad2d((0, 1, 0, 1))\n",
       "            (depthwise_conv): EfficientNetDepthwiseConv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=same, groups=192, bias=False)\n",
       "            (depthwise_norm): BatchNorm2d(192, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (depthwise_act): SiLU()\n",
       "          )\n",
       "          (squeeze_excite): EfficientNetSqueezeExciteLayer(\n",
       "            (squeeze): AdaptiveAvgPool2d(output_size=1)\n",
       "            (reduce): Conv2d(192, 8, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (expand): Conv2d(8, 192, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (act_reduce): SiLU()\n",
       "            (act_expand): Sigmoid()\n",
       "          )\n",
       "          (projection): EfficientNetFinalBlockLayer(\n",
       "            (project_conv): Conv2d(192, 32, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (project_bn): BatchNorm2d(32, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (dropout): Dropout(p=0.02307692307692308, inplace=False)\n",
       "          )\n",
       "        )\n",
       "        (4): EfficientNetBlock(\n",
       "          (expansion): EfficientNetExpansionLayer(\n",
       "            (expand_conv): Conv2d(32, 192, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (expand_bn): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)\n",
       "            (expand_act): SiLU()\n",
       "          )\n",
       "          (depthwise_conv): EfficientNetDepthwiseLayer(\n",
       "            (depthwise_conv_pad): ZeroPad2d((0, 1, 0, 1))\n",
       "            (depthwise_conv): EfficientNetDepthwiseConv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=same, groups=192, bias=False)\n",
       "            (depthwise_norm): BatchNorm2d(192, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (depthwise_act): SiLU()\n",
       "          )\n",
       "          (squeeze_excite): EfficientNetSqueezeExciteLayer(\n",
       "            (squeeze): AdaptiveAvgPool2d(output_size=1)\n",
       "            (reduce): Conv2d(192, 8, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (expand): Conv2d(8, 192, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (act_reduce): SiLU()\n",
       "            (act_expand): Sigmoid()\n",
       "          )\n",
       "          (projection): EfficientNetFinalBlockLayer(\n",
       "            (project_conv): Conv2d(192, 32, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (project_bn): BatchNorm2d(32, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (dropout): Dropout(p=0.03076923076923077, inplace=False)\n",
       "          )\n",
       "        )\n",
       "        (5): EfficientNetBlock(\n",
       "          (expansion): EfficientNetExpansionLayer(\n",
       "            (expand_conv): Conv2d(32, 192, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (expand_bn): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)\n",
       "            (expand_act): SiLU()\n",
       "          )\n",
       "          (depthwise_conv): EfficientNetDepthwiseLayer(\n",
       "            (depthwise_conv_pad): ZeroPad2d((2, 2, 2, 2))\n",
       "            (depthwise_conv): EfficientNetDepthwiseConv2d(192, 192, kernel_size=(5, 5), stride=(2, 2), padding=valid, groups=192, bias=False)\n",
       "            (depthwise_norm): BatchNorm2d(192, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (depthwise_act): SiLU()\n",
       "          )\n",
       "          (squeeze_excite): EfficientNetSqueezeExciteLayer(\n",
       "            (squeeze): AdaptiveAvgPool2d(output_size=1)\n",
       "            (reduce): Conv2d(192, 8, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (expand): Conv2d(8, 192, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (act_reduce): SiLU()\n",
       "            (act_expand): Sigmoid()\n",
       "          )\n",
       "          (projection): EfficientNetFinalBlockLayer(\n",
       "            (project_conv): Conv2d(192, 48, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (project_bn): BatchNorm2d(48, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (dropout): Dropout(p=0.038461538461538464, inplace=False)\n",
       "          )\n",
       "        )\n",
       "        (6): EfficientNetBlock(\n",
       "          (expansion): EfficientNetExpansionLayer(\n",
       "            (expand_conv): Conv2d(48, 288, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (expand_bn): BatchNorm2d(288, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)\n",
       "            (expand_act): SiLU()\n",
       "          )\n",
       "          (depthwise_conv): EfficientNetDepthwiseLayer(\n",
       "            (depthwise_conv_pad): ZeroPad2d((1, 2, 1, 2))\n",
       "            (depthwise_conv): EfficientNetDepthwiseConv2d(288, 288, kernel_size=(5, 5), stride=(1, 1), padding=same, groups=288, bias=False)\n",
       "            (depthwise_norm): BatchNorm2d(288, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (depthwise_act): SiLU()\n",
       "          )\n",
       "          (squeeze_excite): EfficientNetSqueezeExciteLayer(\n",
       "            (squeeze): AdaptiveAvgPool2d(output_size=1)\n",
       "            (reduce): Conv2d(288, 12, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (expand): Conv2d(12, 288, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (act_reduce): SiLU()\n",
       "            (act_expand): Sigmoid()\n",
       "          )\n",
       "          (projection): EfficientNetFinalBlockLayer(\n",
       "            (project_conv): Conv2d(288, 48, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (project_bn): BatchNorm2d(48, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (dropout): Dropout(p=0.04615384615384616, inplace=False)\n",
       "          )\n",
       "        )\n",
       "        (7): EfficientNetBlock(\n",
       "          (expansion): EfficientNetExpansionLayer(\n",
       "            (expand_conv): Conv2d(48, 288, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (expand_bn): BatchNorm2d(288, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)\n",
       "            (expand_act): SiLU()\n",
       "          )\n",
       "          (depthwise_conv): EfficientNetDepthwiseLayer(\n",
       "            (depthwise_conv_pad): ZeroPad2d((1, 2, 1, 2))\n",
       "            (depthwise_conv): EfficientNetDepthwiseConv2d(288, 288, kernel_size=(5, 5), stride=(1, 1), padding=same, groups=288, bias=False)\n",
       "            (depthwise_norm): BatchNorm2d(288, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (depthwise_act): SiLU()\n",
       "          )\n",
       "          (squeeze_excite): EfficientNetSqueezeExciteLayer(\n",
       "            (squeeze): AdaptiveAvgPool2d(output_size=1)\n",
       "            (reduce): Conv2d(288, 12, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (expand): Conv2d(12, 288, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (act_reduce): SiLU()\n",
       "            (act_expand): Sigmoid()\n",
       "          )\n",
       "          (projection): EfficientNetFinalBlockLayer(\n",
       "            (project_conv): Conv2d(288, 48, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (project_bn): BatchNorm2d(48, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (dropout): Dropout(p=0.05384615384615385, inplace=False)\n",
       "          )\n",
       "        )\n",
       "        (8): EfficientNetBlock(\n",
       "          (expansion): EfficientNetExpansionLayer(\n",
       "            (expand_conv): Conv2d(48, 288, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (expand_bn): BatchNorm2d(288, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)\n",
       "            (expand_act): SiLU()\n",
       "          )\n",
       "          (depthwise_conv): EfficientNetDepthwiseLayer(\n",
       "            (depthwise_conv_pad): ZeroPad2d((0, 1, 0, 1))\n",
       "            (depthwise_conv): EfficientNetDepthwiseConv2d(288, 288, kernel_size=(3, 3), stride=(2, 2), padding=valid, groups=288, bias=False)\n",
       "            (depthwise_norm): BatchNorm2d(288, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (depthwise_act): SiLU()\n",
       "          )\n",
       "          (squeeze_excite): EfficientNetSqueezeExciteLayer(\n",
       "            (squeeze): AdaptiveAvgPool2d(output_size=1)\n",
       "            (reduce): Conv2d(288, 12, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (expand): Conv2d(12, 288, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (act_reduce): SiLU()\n",
       "            (act_expand): Sigmoid()\n",
       "          )\n",
       "          (projection): EfficientNetFinalBlockLayer(\n",
       "            (project_conv): Conv2d(288, 96, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (project_bn): BatchNorm2d(96, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (dropout): Dropout(p=0.06153846153846154, inplace=False)\n",
       "          )\n",
       "        )\n",
       "        (9): EfficientNetBlock(\n",
       "          (expansion): EfficientNetExpansionLayer(\n",
       "            (expand_conv): Conv2d(96, 576, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (expand_bn): BatchNorm2d(576, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)\n",
       "            (expand_act): SiLU()\n",
       "          )\n",
       "          (depthwise_conv): EfficientNetDepthwiseLayer(\n",
       "            (depthwise_conv_pad): ZeroPad2d((0, 1, 0, 1))\n",
       "            (depthwise_conv): EfficientNetDepthwiseConv2d(576, 576, kernel_size=(3, 3), stride=(1, 1), padding=same, groups=576, bias=False)\n",
       "            (depthwise_norm): BatchNorm2d(576, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (depthwise_act): SiLU()\n",
       "          )\n",
       "          (squeeze_excite): EfficientNetSqueezeExciteLayer(\n",
       "            (squeeze): AdaptiveAvgPool2d(output_size=1)\n",
       "            (reduce): Conv2d(576, 24, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (expand): Conv2d(24, 576, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (act_reduce): SiLU()\n",
       "            (act_expand): Sigmoid()\n",
       "          )\n",
       "          (projection): EfficientNetFinalBlockLayer(\n",
       "            (project_conv): Conv2d(576, 96, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (project_bn): BatchNorm2d(96, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (dropout): Dropout(p=0.06923076923076923, inplace=False)\n",
       "          )\n",
       "        )\n",
       "        (10): EfficientNetBlock(\n",
       "          (expansion): EfficientNetExpansionLayer(\n",
       "            (expand_conv): Conv2d(96, 576, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (expand_bn): BatchNorm2d(576, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)\n",
       "            (expand_act): SiLU()\n",
       "          )\n",
       "          (depthwise_conv): EfficientNetDepthwiseLayer(\n",
       "            (depthwise_conv_pad): ZeroPad2d((0, 1, 0, 1))\n",
       "            (depthwise_conv): EfficientNetDepthwiseConv2d(576, 576, kernel_size=(3, 3), stride=(1, 1), padding=same, groups=576, bias=False)\n",
       "            (depthwise_norm): BatchNorm2d(576, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (depthwise_act): SiLU()\n",
       "          )\n",
       "          (squeeze_excite): EfficientNetSqueezeExciteLayer(\n",
       "            (squeeze): AdaptiveAvgPool2d(output_size=1)\n",
       "            (reduce): Conv2d(576, 24, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (expand): Conv2d(24, 576, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (act_reduce): SiLU()\n",
       "            (act_expand): Sigmoid()\n",
       "          )\n",
       "          (projection): EfficientNetFinalBlockLayer(\n",
       "            (project_conv): Conv2d(576, 96, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (project_bn): BatchNorm2d(96, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (dropout): Dropout(p=0.07692307692307693, inplace=False)\n",
       "          )\n",
       "        )\n",
       "        (11): EfficientNetBlock(\n",
       "          (expansion): EfficientNetExpansionLayer(\n",
       "            (expand_conv): Conv2d(96, 576, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (expand_bn): BatchNorm2d(576, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)\n",
       "            (expand_act): SiLU()\n",
       "          )\n",
       "          (depthwise_conv): EfficientNetDepthwiseLayer(\n",
       "            (depthwise_conv_pad): ZeroPad2d((0, 1, 0, 1))\n",
       "            (depthwise_conv): EfficientNetDepthwiseConv2d(576, 576, kernel_size=(3, 3), stride=(1, 1), padding=same, groups=576, bias=False)\n",
       "            (depthwise_norm): BatchNorm2d(576, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (depthwise_act): SiLU()\n",
       "          )\n",
       "          (squeeze_excite): EfficientNetSqueezeExciteLayer(\n",
       "            (squeeze): AdaptiveAvgPool2d(output_size=1)\n",
       "            (reduce): Conv2d(576, 24, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (expand): Conv2d(24, 576, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (act_reduce): SiLU()\n",
       "            (act_expand): Sigmoid()\n",
       "          )\n",
       "          (projection): EfficientNetFinalBlockLayer(\n",
       "            (project_conv): Conv2d(576, 96, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (project_bn): BatchNorm2d(96, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (dropout): Dropout(p=0.08461538461538462, inplace=False)\n",
       "          )\n",
       "        )\n",
       "        (12): EfficientNetBlock(\n",
       "          (expansion): EfficientNetExpansionLayer(\n",
       "            (expand_conv): Conv2d(96, 576, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (expand_bn): BatchNorm2d(576, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)\n",
       "            (expand_act): SiLU()\n",
       "          )\n",
       "          (depthwise_conv): EfficientNetDepthwiseLayer(\n",
       "            (depthwise_conv_pad): ZeroPad2d((0, 1, 0, 1))\n",
       "            (depthwise_conv): EfficientNetDepthwiseConv2d(576, 576, kernel_size=(3, 3), stride=(1, 1), padding=same, groups=576, bias=False)\n",
       "            (depthwise_norm): BatchNorm2d(576, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (depthwise_act): SiLU()\n",
       "          )\n",
       "          (squeeze_excite): EfficientNetSqueezeExciteLayer(\n",
       "            (squeeze): AdaptiveAvgPool2d(output_size=1)\n",
       "            (reduce): Conv2d(576, 24, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (expand): Conv2d(24, 576, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (act_reduce): SiLU()\n",
       "            (act_expand): Sigmoid()\n",
       "          )\n",
       "          (projection): EfficientNetFinalBlockLayer(\n",
       "            (project_conv): Conv2d(576, 96, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (project_bn): BatchNorm2d(96, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (dropout): Dropout(p=0.09230769230769233, inplace=False)\n",
       "          )\n",
       "        )\n",
       "        (13): EfficientNetBlock(\n",
       "          (expansion): EfficientNetExpansionLayer(\n",
       "            (expand_conv): Conv2d(96, 576, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (expand_bn): BatchNorm2d(576, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)\n",
       "            (expand_act): SiLU()\n",
       "          )\n",
       "          (depthwise_conv): EfficientNetDepthwiseLayer(\n",
       "            (depthwise_conv_pad): ZeroPad2d((1, 2, 1, 2))\n",
       "            (depthwise_conv): EfficientNetDepthwiseConv2d(576, 576, kernel_size=(5, 5), stride=(1, 1), padding=same, groups=576, bias=False)\n",
       "            (depthwise_norm): BatchNorm2d(576, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (depthwise_act): SiLU()\n",
       "          )\n",
       "          (squeeze_excite): EfficientNetSqueezeExciteLayer(\n",
       "            (squeeze): AdaptiveAvgPool2d(output_size=1)\n",
       "            (reduce): Conv2d(576, 24, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (expand): Conv2d(24, 576, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (act_reduce): SiLU()\n",
       "            (act_expand): Sigmoid()\n",
       "          )\n",
       "          (projection): EfficientNetFinalBlockLayer(\n",
       "            (project_conv): Conv2d(576, 136, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (project_bn): BatchNorm2d(136, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (dropout): Dropout(p=0.1, inplace=False)\n",
       "          )\n",
       "        )\n",
       "        (14): EfficientNetBlock(\n",
       "          (expansion): EfficientNetExpansionLayer(\n",
       "            (expand_conv): Conv2d(136, 816, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (expand_bn): BatchNorm2d(816, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)\n",
       "            (expand_act): SiLU()\n",
       "          )\n",
       "          (depthwise_conv): EfficientNetDepthwiseLayer(\n",
       "            (depthwise_conv_pad): ZeroPad2d((1, 2, 1, 2))\n",
       "            (depthwise_conv): EfficientNetDepthwiseConv2d(816, 816, kernel_size=(5, 5), stride=(1, 1), padding=same, groups=816, bias=False)\n",
       "            (depthwise_norm): BatchNorm2d(816, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (depthwise_act): SiLU()\n",
       "          )\n",
       "          (squeeze_excite): EfficientNetSqueezeExciteLayer(\n",
       "            (squeeze): AdaptiveAvgPool2d(output_size=1)\n",
       "            (reduce): Conv2d(816, 34, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (expand): Conv2d(34, 816, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (act_reduce): SiLU()\n",
       "            (act_expand): Sigmoid()\n",
       "          )\n",
       "          (projection): EfficientNetFinalBlockLayer(\n",
       "            (project_conv): Conv2d(816, 136, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (project_bn): BatchNorm2d(136, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (dropout): Dropout(p=0.1076923076923077, inplace=False)\n",
       "          )\n",
       "        )\n",
       "        (15): EfficientNetBlock(\n",
       "          (expansion): EfficientNetExpansionLayer(\n",
       "            (expand_conv): Conv2d(136, 816, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (expand_bn): BatchNorm2d(816, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)\n",
       "            (expand_act): SiLU()\n",
       "          )\n",
       "          (depthwise_conv): EfficientNetDepthwiseLayer(\n",
       "            (depthwise_conv_pad): ZeroPad2d((1, 2, 1, 2))\n",
       "            (depthwise_conv): EfficientNetDepthwiseConv2d(816, 816, kernel_size=(5, 5), stride=(1, 1), padding=same, groups=816, bias=False)\n",
       "            (depthwise_norm): BatchNorm2d(816, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (depthwise_act): SiLU()\n",
       "          )\n",
       "          (squeeze_excite): EfficientNetSqueezeExciteLayer(\n",
       "            (squeeze): AdaptiveAvgPool2d(output_size=1)\n",
       "            (reduce): Conv2d(816, 34, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (expand): Conv2d(34, 816, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (act_reduce): SiLU()\n",
       "            (act_expand): Sigmoid()\n",
       "          )\n",
       "          (projection): EfficientNetFinalBlockLayer(\n",
       "            (project_conv): Conv2d(816, 136, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (project_bn): BatchNorm2d(136, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (dropout): Dropout(p=0.11538461538461539, inplace=False)\n",
       "          )\n",
       "        )\n",
       "        (16): EfficientNetBlock(\n",
       "          (expansion): EfficientNetExpansionLayer(\n",
       "            (expand_conv): Conv2d(136, 816, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (expand_bn): BatchNorm2d(816, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)\n",
       "            (expand_act): SiLU()\n",
       "          )\n",
       "          (depthwise_conv): EfficientNetDepthwiseLayer(\n",
       "            (depthwise_conv_pad): ZeroPad2d((1, 2, 1, 2))\n",
       "            (depthwise_conv): EfficientNetDepthwiseConv2d(816, 816, kernel_size=(5, 5), stride=(1, 1), padding=same, groups=816, bias=False)\n",
       "            (depthwise_norm): BatchNorm2d(816, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (depthwise_act): SiLU()\n",
       "          )\n",
       "          (squeeze_excite): EfficientNetSqueezeExciteLayer(\n",
       "            (squeeze): AdaptiveAvgPool2d(output_size=1)\n",
       "            (reduce): Conv2d(816, 34, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (expand): Conv2d(34, 816, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (act_reduce): SiLU()\n",
       "            (act_expand): Sigmoid()\n",
       "          )\n",
       "          (projection): EfficientNetFinalBlockLayer(\n",
       "            (project_conv): Conv2d(816, 136, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (project_bn): BatchNorm2d(136, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (dropout): Dropout(p=0.12307692307692308, inplace=False)\n",
       "          )\n",
       "        )\n",
       "        (17): EfficientNetBlock(\n",
       "          (expansion): EfficientNetExpansionLayer(\n",
       "            (expand_conv): Conv2d(136, 816, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (expand_bn): BatchNorm2d(816, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)\n",
       "            (expand_act): SiLU()\n",
       "          )\n",
       "          (depthwise_conv): EfficientNetDepthwiseLayer(\n",
       "            (depthwise_conv_pad): ZeroPad2d((1, 2, 1, 2))\n",
       "            (depthwise_conv): EfficientNetDepthwiseConv2d(816, 816, kernel_size=(5, 5), stride=(1, 1), padding=same, groups=816, bias=False)\n",
       "            (depthwise_norm): BatchNorm2d(816, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (depthwise_act): SiLU()\n",
       "          )\n",
       "          (squeeze_excite): EfficientNetSqueezeExciteLayer(\n",
       "            (squeeze): AdaptiveAvgPool2d(output_size=1)\n",
       "            (reduce): Conv2d(816, 34, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (expand): Conv2d(34, 816, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (act_reduce): SiLU()\n",
       "            (act_expand): Sigmoid()\n",
       "          )\n",
       "          (projection): EfficientNetFinalBlockLayer(\n",
       "            (project_conv): Conv2d(816, 136, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (project_bn): BatchNorm2d(136, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (dropout): Dropout(p=0.13076923076923078, inplace=False)\n",
       "          )\n",
       "        )\n",
       "        (18): EfficientNetBlock(\n",
       "          (expansion): EfficientNetExpansionLayer(\n",
       "            (expand_conv): Conv2d(136, 816, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (expand_bn): BatchNorm2d(816, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)\n",
       "            (expand_act): SiLU()\n",
       "          )\n",
       "          (depthwise_conv): EfficientNetDepthwiseLayer(\n",
       "            (depthwise_conv_pad): ZeroPad2d((2, 2, 2, 2))\n",
       "            (depthwise_conv): EfficientNetDepthwiseConv2d(816, 816, kernel_size=(5, 5), stride=(2, 2), padding=valid, groups=816, bias=False)\n",
       "            (depthwise_norm): BatchNorm2d(816, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (depthwise_act): SiLU()\n",
       "          )\n",
       "          (squeeze_excite): EfficientNetSqueezeExciteLayer(\n",
       "            (squeeze): AdaptiveAvgPool2d(output_size=1)\n",
       "            (reduce): Conv2d(816, 34, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (expand): Conv2d(34, 816, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (act_reduce): SiLU()\n",
       "            (act_expand): Sigmoid()\n",
       "          )\n",
       "          (projection): EfficientNetFinalBlockLayer(\n",
       "            (project_conv): Conv2d(816, 232, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (project_bn): BatchNorm2d(232, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (dropout): Dropout(p=0.13846153846153847, inplace=False)\n",
       "          )\n",
       "        )\n",
       "        (19): EfficientNetBlock(\n",
       "          (expansion): EfficientNetExpansionLayer(\n",
       "            (expand_conv): Conv2d(232, 1392, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (expand_bn): BatchNorm2d(1392, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)\n",
       "            (expand_act): SiLU()\n",
       "          )\n",
       "          (depthwise_conv): EfficientNetDepthwiseLayer(\n",
       "            (depthwise_conv_pad): ZeroPad2d((1, 2, 1, 2))\n",
       "            (depthwise_conv): EfficientNetDepthwiseConv2d(1392, 1392, kernel_size=(5, 5), stride=(1, 1), padding=same, groups=1392, bias=False)\n",
       "            (depthwise_norm): BatchNorm2d(1392, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (depthwise_act): SiLU()\n",
       "          )\n",
       "          (squeeze_excite): EfficientNetSqueezeExciteLayer(\n",
       "            (squeeze): AdaptiveAvgPool2d(output_size=1)\n",
       "            (reduce): Conv2d(1392, 58, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (expand): Conv2d(58, 1392, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (act_reduce): SiLU()\n",
       "            (act_expand): Sigmoid()\n",
       "          )\n",
       "          (projection): EfficientNetFinalBlockLayer(\n",
       "            (project_conv): Conv2d(1392, 232, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (project_bn): BatchNorm2d(232, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (dropout): Dropout(p=0.14615384615384616, inplace=False)\n",
       "          )\n",
       "        )\n",
       "        (20): EfficientNetBlock(\n",
       "          (expansion): EfficientNetExpansionLayer(\n",
       "            (expand_conv): Conv2d(232, 1392, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (expand_bn): BatchNorm2d(1392, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)\n",
       "            (expand_act): SiLU()\n",
       "          )\n",
       "          (depthwise_conv): EfficientNetDepthwiseLayer(\n",
       "            (depthwise_conv_pad): ZeroPad2d((1, 2, 1, 2))\n",
       "            (depthwise_conv): EfficientNetDepthwiseConv2d(1392, 1392, kernel_size=(5, 5), stride=(1, 1), padding=same, groups=1392, bias=False)\n",
       "            (depthwise_norm): BatchNorm2d(1392, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (depthwise_act): SiLU()\n",
       "          )\n",
       "          (squeeze_excite): EfficientNetSqueezeExciteLayer(\n",
       "            (squeeze): AdaptiveAvgPool2d(output_size=1)\n",
       "            (reduce): Conv2d(1392, 58, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (expand): Conv2d(58, 1392, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (act_reduce): SiLU()\n",
       "            (act_expand): Sigmoid()\n",
       "          )\n",
       "          (projection): EfficientNetFinalBlockLayer(\n",
       "            (project_conv): Conv2d(1392, 232, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (project_bn): BatchNorm2d(232, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (dropout): Dropout(p=0.15384615384615385, inplace=False)\n",
       "          )\n",
       "        )\n",
       "        (21): EfficientNetBlock(\n",
       "          (expansion): EfficientNetExpansionLayer(\n",
       "            (expand_conv): Conv2d(232, 1392, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (expand_bn): BatchNorm2d(1392, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)\n",
       "            (expand_act): SiLU()\n",
       "          )\n",
       "          (depthwise_conv): EfficientNetDepthwiseLayer(\n",
       "            (depthwise_conv_pad): ZeroPad2d((1, 2, 1, 2))\n",
       "            (depthwise_conv): EfficientNetDepthwiseConv2d(1392, 1392, kernel_size=(5, 5), stride=(1, 1), padding=same, groups=1392, bias=False)\n",
       "            (depthwise_norm): BatchNorm2d(1392, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (depthwise_act): SiLU()\n",
       "          )\n",
       "          (squeeze_excite): EfficientNetSqueezeExciteLayer(\n",
       "            (squeeze): AdaptiveAvgPool2d(output_size=1)\n",
       "            (reduce): Conv2d(1392, 58, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (expand): Conv2d(58, 1392, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (act_reduce): SiLU()\n",
       "            (act_expand): Sigmoid()\n",
       "          )\n",
       "          (projection): EfficientNetFinalBlockLayer(\n",
       "            (project_conv): Conv2d(1392, 232, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (project_bn): BatchNorm2d(232, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (dropout): Dropout(p=0.16153846153846155, inplace=False)\n",
       "          )\n",
       "        )\n",
       "        (22): EfficientNetBlock(\n",
       "          (expansion): EfficientNetExpansionLayer(\n",
       "            (expand_conv): Conv2d(232, 1392, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (expand_bn): BatchNorm2d(1392, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)\n",
       "            (expand_act): SiLU()\n",
       "          )\n",
       "          (depthwise_conv): EfficientNetDepthwiseLayer(\n",
       "            (depthwise_conv_pad): ZeroPad2d((1, 2, 1, 2))\n",
       "            (depthwise_conv): EfficientNetDepthwiseConv2d(1392, 1392, kernel_size=(5, 5), stride=(1, 1), padding=same, groups=1392, bias=False)\n",
       "            (depthwise_norm): BatchNorm2d(1392, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (depthwise_act): SiLU()\n",
       "          )\n",
       "          (squeeze_excite): EfficientNetSqueezeExciteLayer(\n",
       "            (squeeze): AdaptiveAvgPool2d(output_size=1)\n",
       "            (reduce): Conv2d(1392, 58, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (expand): Conv2d(58, 1392, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (act_reduce): SiLU()\n",
       "            (act_expand): Sigmoid()\n",
       "          )\n",
       "          (projection): EfficientNetFinalBlockLayer(\n",
       "            (project_conv): Conv2d(1392, 232, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (project_bn): BatchNorm2d(232, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (dropout): Dropout(p=0.16923076923076924, inplace=False)\n",
       "          )\n",
       "        )\n",
       "        (23): EfficientNetBlock(\n",
       "          (expansion): EfficientNetExpansionLayer(\n",
       "            (expand_conv): Conv2d(232, 1392, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (expand_bn): BatchNorm2d(1392, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)\n",
       "            (expand_act): SiLU()\n",
       "          )\n",
       "          (depthwise_conv): EfficientNetDepthwiseLayer(\n",
       "            (depthwise_conv_pad): ZeroPad2d((1, 2, 1, 2))\n",
       "            (depthwise_conv): EfficientNetDepthwiseConv2d(1392, 1392, kernel_size=(5, 5), stride=(1, 1), padding=same, groups=1392, bias=False)\n",
       "            (depthwise_norm): BatchNorm2d(1392, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (depthwise_act): SiLU()\n",
       "          )\n",
       "          (squeeze_excite): EfficientNetSqueezeExciteLayer(\n",
       "            (squeeze): AdaptiveAvgPool2d(output_size=1)\n",
       "            (reduce): Conv2d(1392, 58, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (expand): Conv2d(58, 1392, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (act_reduce): SiLU()\n",
       "            (act_expand): Sigmoid()\n",
       "          )\n",
       "          (projection): EfficientNetFinalBlockLayer(\n",
       "            (project_conv): Conv2d(1392, 232, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (project_bn): BatchNorm2d(232, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (dropout): Dropout(p=0.17692307692307693, inplace=False)\n",
       "          )\n",
       "        )\n",
       "        (24): EfficientNetBlock(\n",
       "          (expansion): EfficientNetExpansionLayer(\n",
       "            (expand_conv): Conv2d(232, 1392, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (expand_bn): BatchNorm2d(1392, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)\n",
       "            (expand_act): SiLU()\n",
       "          )\n",
       "          (depthwise_conv): EfficientNetDepthwiseLayer(\n",
       "            (depthwise_conv_pad): ZeroPad2d((0, 1, 0, 1))\n",
       "            (depthwise_conv): EfficientNetDepthwiseConv2d(1392, 1392, kernel_size=(3, 3), stride=(1, 1), padding=same, groups=1392, bias=False)\n",
       "            (depthwise_norm): BatchNorm2d(1392, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (depthwise_act): SiLU()\n",
       "          )\n",
       "          (squeeze_excite): EfficientNetSqueezeExciteLayer(\n",
       "            (squeeze): AdaptiveAvgPool2d(output_size=1)\n",
       "            (reduce): Conv2d(1392, 58, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (expand): Conv2d(58, 1392, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (act_reduce): SiLU()\n",
       "            (act_expand): Sigmoid()\n",
       "          )\n",
       "          (projection): EfficientNetFinalBlockLayer(\n",
       "            (project_conv): Conv2d(1392, 384, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (project_bn): BatchNorm2d(384, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (dropout): Dropout(p=0.18461538461538465, inplace=False)\n",
       "          )\n",
       "        )\n",
       "        (25): EfficientNetBlock(\n",
       "          (expansion): EfficientNetExpansionLayer(\n",
       "            (expand_conv): Conv2d(384, 2304, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (expand_bn): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)\n",
       "            (expand_act): SiLU()\n",
       "          )\n",
       "          (depthwise_conv): EfficientNetDepthwiseLayer(\n",
       "            (depthwise_conv_pad): ZeroPad2d((0, 1, 0, 1))\n",
       "            (depthwise_conv): EfficientNetDepthwiseConv2d(2304, 2304, kernel_size=(3, 3), stride=(1, 1), padding=same, groups=2304, bias=False)\n",
       "            (depthwise_norm): BatchNorm2d(2304, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (depthwise_act): SiLU()\n",
       "          )\n",
       "          (squeeze_excite): EfficientNetSqueezeExciteLayer(\n",
       "            (squeeze): AdaptiveAvgPool2d(output_size=1)\n",
       "            (reduce): Conv2d(2304, 96, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (expand): Conv2d(96, 2304, kernel_size=(1, 1), stride=(1, 1), padding=same)\n",
       "            (act_reduce): SiLU()\n",
       "            (act_expand): Sigmoid()\n",
       "          )\n",
       "          (projection): EfficientNetFinalBlockLayer(\n",
       "            (project_conv): Conv2d(2304, 384, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "            (project_bn): BatchNorm2d(384, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "            (dropout): Dropout(p=0.19230769230769232, inplace=False)\n",
       "          )\n",
       "        )\n",
       "      )\n",
       "      (top_conv): Conv2d(384, 1536, kernel_size=(1, 1), stride=(1, 1), padding=same, bias=False)\n",
       "      (top_bn): BatchNorm2d(1536, eps=0.001, momentum=0.99, affine=True, track_running_stats=True)\n",
       "      (top_activation): SiLU()\n",
       "    )\n",
       "    (pooler): AvgPool2d(kernel_size=1536, stride=1536, padding=0)\n",
       "  )\n",
       "  (dropout): Dropout(p=0.3, inplace=False)\n",
       "  (classifier): Linear(in_features=1536, out_features=3872, bias=True)\n",
       ")"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "ArtifactClassification",
   "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": "3.10.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}