Upload Zouhayer_v3.ipynb

Zouhayer_v3.ipynb

@@ -0,0 +1,2116 @@
+{
+ "nbformat": 4,
+ "nbformat_minor": 0,
+ "metadata": {
+  "colab": {
+   "provenance": [],
+   "gpuType": "T4"
+  },
+  "kernelspec": {
+   "name": "python3",
+   "display_name": "Python 3"
+  },
+  "language_info": {
+   "name": "python"
+  },
+  "accelerator": "GPU"
+ },
+ "cells": [
+  {
+   "cell_type": "code",
+   "source": [
+    "from google.colab import drive\n",
+    "drive.mount('/content/drive')"
+   ],
+   "metadata": {
+    "colab": {
+     "base_uri": "https://localhost:8080/"
+    },
+    "id": "1JHSkdJSJLlV",
+    "outputId": "4bc51836-8266-4a47-d50c-a3d7b7abdce0"
+   },
+   "execution_count": 1,
+   "outputs": [
+    {
+     "output_type": "stream",
+     "name": "stdout",
+     "text": [
+      "Mounted at /content/drive\n"
+     ]
+    }
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "metadata": {
+    "colab": {
+     "base_uri": "https://localhost:8080/"
+    },
+    "id": "pYiI4zfBztum",
+    "outputId": "cc5f1000-52d0-433f-a44a-b6526a3daf4f"
+   },
+   "outputs": [
+    {
+     "output_type": "stream",
+     "name": "stdout",
+     "text": [
+      "Collecting fasttext\n",
+      "  Downloading fasttext-0.9.3.tar.gz (73 kB)\n",
+      "\u001b[?25l     \u001b[90m\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u001b[0m \u001b[32m0.0/73.4 kB\u001b[0m \u001b[31m?\u001b[0m eta \u001b[36m-:--:--\u001b[0m\r\u001b[2K     \u001b[90m\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u001b[0m \u001b[32m73.4/73.4 kB\u001b[0m \u001b[31m4.7 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
+      "\u001b[?25h  Installing build dependencies ... \u001b[?25l\u001b[?25hdone\n",
+      "  Getting requirements to build wheel ... \u001b[?25l\u001b[?25hdone\n",
+      "  Preparing metadata (pyproject.toml) ... \u001b[?25l\u001b[?25hdone\n",
+      "Collecting pybind11>=2.2 (from fasttext)\n",
+      "  Using cached pybind11-3.0.3-py3-none-any.whl.metadata (10 kB)\n",
+      "Requirement already satisfied: setuptools>=0.7.0 in /usr/local/lib/python3.12/dist-packages (from fasttext) (75.2.0)\n",
+      "Requirement already satisfied: numpy in /usr/local/lib/python3.12/dist-packages (from fasttext) (2.0.2)\n",
+      "Using cached pybind11-3.0.3-py3-none-any.whl (313 kB)\n",
+      "Building wheels for collected packages: fasttext\n",
+      "  Building wheel for fasttext (pyproject.toml) ... \u001b[?25l\u001b[?25hdone\n",
+      "  Created wheel for fasttext: filename=fasttext-0.9.3-cp312-cp312-linux_x86_64.whl size=4653979 sha256=a6c11b55dfb9ca99d6c2ad8c026d2d0bcee4cd11dbfb8cd9864ca7de1260b569\n",
+      "  Stored in directory: /root/.cache/pip/wheels/20/27/95/a7baf1b435f1cbde017cabdf1e9688526d2b0e929255a359c6\n",
+      "Successfully built fasttext\n",
+      "Installing collected packages: pybind11, fasttext\n",
+      "Successfully installed fasttext-0.9.3 pybind11-3.0.3\n"
+     ]
+    }
+   ],
+   "source": [
+    "!pip install fasttext"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "source": [
+    "import pandas as pd\n",
+    "import numpy as np\n",
+    "corpus_full = pd.read_csv(\"/content/ArabiziProfanityDatasetByMotaz.csv\").astype(str)\n",
+    "corpus_text = np.array(corpus_full[\"text\"].tolist(), dtype=str)"
+   ],
+   "metadata": {
+    "id": "R0bAuNkz2HDx"
+   },
+   "execution_count": 3,
+   "outputs": []
+  },
+  {
+   "cell_type": "code",
+   "source": [
+    "import fasttext\n",
+    "import pandas as pd\n",
+    "import numpy as np\n",
+    "import os\n",
+    "# Ensure data is loaded\n",
+    "corpus_full = pd.read_csv(\"/content/ArabiziProfanityDatasetByMotaz.csv\").astype(str)\n",
+    "corpus_text = np.array(corpus_full[\"text\"].tolist(), dtype=str)\n",
+    "\n",
+    "# Save the corpus to a text file as required by fastText\n",
+    "with open('corpus.txt', 'w', encoding='utf-8') as f:\n",
+    "    for line in corpus_text:\n",
+    "        f.write(str(line) + '\\n')\n",
+    "\n",
+    "# Train the model with verbose=2 to see progress\n",
+    "print(\"Starting training...\")\n",
+    "model = fasttext.train_unsupervised(input=\"corpus.txt\", model=\"skipgram\", lr=0.05, dim=256, minn=2, maxn=8, wordNgrams=3, ws=5, epoch=20, verbose=2)\n",
+    "model.save_model(\"TunisianEmbeddings.ftz\")\n",
+    "print(\"Model saved as cc.ftz\")"
+   ],
+   "metadata": {
+    "colab": {
+     "base_uri": "https://localhost:8080/"
+    },
+    "id": "cGatmAbG1SE4",
+    "outputId": "7479bde2-2edb-41af-d7cf-dfca701e5fdb"
+   },
+   "execution_count": null,
+   "outputs": [
+    {
+     "output_type": "stream",
+     "name": "stdout",
+     "text": [
+      "Starting training...\n",
+      "Model saved as cc.ftz\n"
+     ]
+    }
+   ]
+  },
+  {
+   "cell_type": "code",
+   "source": [
+    "# Copies the file to the root of your Google Drive\n",
+    "!cp /content/TunisianEmbeddings.ftz /content/drive/MyDrive/"
+   ],
+   "metadata": {
+    "id": "KGkrmaYfJ021"
+   },
+   "execution_count": null,
+   "outputs": []
+  },
+  {
+   "cell_type": "code",
+   "source": [
+    "import torch"
+   ],
+   "metadata": {
+    "id": "xWdg88mW8Nn9"
+   },
+   "execution_count": null,
+   "outputs": []
+  },
+  {
+   "cell_type": "code",
+   "source": [
+    "import torch.nn as nn"
+   ],
+   "metadata": {
+    "id": "IngqsV4b8QTL"
+   },
+   "execution_count": null,
+   "outputs": []
+  },
+  {
+   "cell_type": "code",
+   "source": [
+    "import torch\n",
+    "import torch.nn as nn\n",
+    "\n",
+    "# \u2500\u2500 LiGRU cell \u2014 LayerNorm, no BN \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\n",
+    "class LiGRUCell(nn.Module):\n",
+    "    def __init__(self, input_size, hidden_size, dropout=0.0):\n",
+    "        super().__init__()\n",
+    "        self.hidden_size = hidden_size\n",
+    "        self.Wz = nn.Linear(input_size + hidden_size, hidden_size)\n",
+    "        self.Wh = nn.Linear(input_size + hidden_size, hidden_size)\n",
+    "        self.ln_h = nn.LayerNorm(hidden_size)\n",
+    "        self.drop = nn.Dropout(dropout)\n",
+    "\n",
+    "    def forward(self, x, h):\n",
+    "        combined = torch.cat([x, h], dim=-1)\n",
+    "        z = torch.sigmoid(self.Wz(combined))\n",
+    "        h_candidate = torch.relu(self.ln_h(self.Wh(combined)))\n",
+    "        h_candidate = self.drop(h_candidate)\n",
+    "        return (1 - z) * h + z * h_candidate\n",
+    "\n",
+    "\n",
+    "# \u2500\u2500 Bidirectional LiGRU \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\n",
+    "class BiLiGRU(nn.Module):\n",
+    "    def __init__(self, input_size, hidden_size, dropout=0.4):\n",
+    "        super().__init__()\n",
+    "        self.hidden_size = hidden_size\n",
+    "        self.fwd_cell = LiGRUCell(input_size, hidden_size, dropout=dropout)\n",
+    "        self.bwd_cell = LiGRUCell(input_size, hidden_size, dropout=dropout)\n",
+    "        self.out_drop = nn.Dropout(dropout)\n",
+    "\n",
+    "    def forward(self, x):\n",
+    "        batch, seq_len, _ = x.shape\n",
+    "        h_fwd = torch.zeros(batch, self.hidden_size, device=x.device)\n",
+    "        h_bwd = torch.zeros(batch, self.hidden_size, device=x.device)\n",
+    "        fwd_out, bwd_out = [], []\n",
+    "        for t in range(seq_len):\n",
+    "            h_fwd = self.fwd_cell(x[:, t, :], h_fwd)\n",
+    "            h_bwd = self.bwd_cell(x[:, seq_len - 1 - t, :], h_bwd)\n",
+    "            fwd_out.append(h_fwd)\n",
+    "            bwd_out.append(h_bwd)\n",
+    "        bwd_out.reverse()\n",
+    "        out = torch.stack(\n",
+    "            [torch.cat([f, b], dim=-1) for f, b in zip(fwd_out, bwd_out)], dim=1\n",
+    "        )  # [B, T, H*2]\n",
+    "        return self.out_drop(out)\n",
+    "\n",
+    "\n",
+    "# \u2500\u2500 Asfour v3 \u2014 anti-overfitting redesign \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\n",
+    "class Asfour(nn.Module):\n",
+    "    \"\"\"\n",
+    "    v3 vs v2 changes (all anti-overfitting):\n",
+    "      - REMOVED RoPE:          overkill for \u226430-token sequences; added memorisation risk\n",
+    "      - REMOVED CNN block:     too many stacked components for dataset size\n",
+    "      - hidden_size 64 \u2192 48:   further capacity cut (~67% fewer total params vs v2)\n",
+    "      - dropout 0.25 \u2192 0.40:   meaningful regularisation throughout the recurrent path\n",
+    "      - max+avg pooling \u2192 learned attention pooling: fewer params, differentiable focus\n",
+    "      - head simplified to Dropout \u2192 Linear: removes extra dense layer that was overfitting\n",
+    "    \"\"\"\n",
+    "    def __init__(self, input_dim=256, hidden_size=48, num_classes=1, dropout=0.4):\n",
+    "        super().__init__()\n",
+    "\n",
+    "        # Input projection with built-in regularisation\n",
+    "        self.proj = nn.Sequential(\n",
+    "            nn.Linear(input_dim, hidden_size),\n",
+    "            nn.LayerNorm(hidden_size),\n",
+    "            nn.ReLU(),\n",
+    "            nn.Dropout(dropout),\n",
+    "        )\n",
+    "\n",
+    "        self.bigru = BiLiGRU(\n",
+    "            input_size=hidden_size,\n",
+    "            hidden_size=hidden_size,\n",
+    "            dropout=dropout,\n",
+    "        )  # output: [B, T, hidden_size*2]\n",
+    "\n",
+    "        # Learned attention pooling \u2014 collapses sequence without bloating params\n",
+    "        self.attn = nn.Linear(hidden_size * 2, 1, bias=False)\n",
+    "\n",
+    "        # Minimal head: single linear after dropout\n",
+    "        self.head = nn.Sequential(\n",
+    "            nn.Dropout(dropout),\n",
+    "            nn.Linear(hidden_size * 2, num_classes),\n",
+    "        )\n",
+    "\n",
+    "    def forward(self, x):\n",
+    "        # x: [B, T, 256]\n",
+    "        x = self.proj(x)                           # [B, T, H]\n",
+    "        x = self.bigru(x)                          # [B, T, H*2]\n",
+    "\n",
+    "        # Attention pooling\n",
+    "        w = torch.softmax(self.attn(x), dim=1)     # [B, T, 1]\n",
+    "        x = (x * w).sum(dim=1)                     # [B, H*2]\n",
+    "\n",
+    "        return self.head(x)                        # [B, 1]\n"
+   ],
+   "metadata": {
+    "id": "KPoh1f6y8SPY"
+   },
+   "execution_count": 4,
+   "outputs": []
+  },
+  {
+   "cell_type": "code",
+   "source": [
+    "import fasttext\n",
+    "\n",
+    "model = fasttext.load_model(\"/content/drive/MyDrive/cc.ftz\")"
+   ],
+   "metadata": {
+    "id": "2h4pWkfxE_yS"
+   },
+   "execution_count": 5,
+   "outputs": []
+  },
+  {
+   "cell_type": "code",
+   "metadata": {
+    "id": "0c95f755"
+   },
+   "source": [
+    "i",
+    "m",
+    "p",
+    "o",
+    "r",
+    "t",
+    " ",
+    "t",
+    "o",
+    "r",
+    "c",
+    "h",
+    "\n",
+    "f",
+    "r",
+    "o",
+    "m",
+    " ",
+    "t",
+    "o",
+    "r",
+    "c",
+    "h",
+    ".",
+    "u",
+    "t",
+    "i",
+    "l",
+    "s",
+    ".",
+    "d",
+    "a",
+    "t",
+    "a",
+    " ",
+    "i",
+    "m",
+    "p",
+    "o",
+    "r",
+    "t",
+    " ",
+    "D",
+    "a",
+    "t",
+    "a",
+    "s",
+    "e",
+    "t",
+    ",",
+    " ",
+    "D",
+    "a",
+    "t",
+    "a",
+    "L",
+    "o",
+    "a",
+    "d",
+    "e",
+    "r",
+    "\n",
+    "i",
+    "m",
+    "p",
+    "o",
+    "r",
+    "t",
+    " ",
+    "n",
+    "u",
+    "m",
+    "p",
+    "y",
+    " ",
+    "a",
+    "s",
+    " ",
+    "n",
+    "p",
+    "\n",
+    "\n",
+    "c",
+    "l",
+    "a",
+    "s",
+    "s",
+    " ",
+    "T",
+    "u",
+    "n",
+    "i",
+    "s",
+    "i",
+    "a",
+    "n",
+    "D",
+    "a",
+    "t",
+    "a",
+    "s",
+    "e",
+    "t",
+    "(",
+    "D",
+    "a",
+    "t",
+    "a",
+    "s",
+    "e",
+    "t",
+    ")",
+    ":",
+    "\n",
+    " ",
+    " ",
+    " ",
+    " ",
+    "d",
+    "e",
+    "f",
+    " ",
+    "_",
+    "_",
+    "i",
+    "n",
+    "i",
+    "t",
+    "_",
+    "_",
+    "(",
+    "s",
+    "e",
+    "l",
+    "f",
+    ",",
+    " ",
+    "t",
+    "e",
+    "x",
+    "t",
+    "s",
+    ",",
+    " ",
+    "l",
+    "a",
+    "b",
+    "e",
+    "l",
+    "s",
+    ",",
+    " ",
+    "f",
+    "a",
+    "s",
+    "t",
+    "t",
+    "e",
+    "x",
+    "t",
+    "_",
+    "m",
+    "o",
+    "d",
+    "e",
+    "l",
+    ",",
+    " ",
+    "m",
+    "a",
+    "x",
+    "_",
+    "l",
+    "e",
+    "n",
+    "=",
+    "3",
+    "0",
+    ")",
+    ":",
+    "\n",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    "s",
+    "e",
+    "l",
+    "f",
+    ".",
+    "t",
+    "e",
+    "x",
+    "t",
+    "s",
+    " ",
+    "=",
+    " ",
+    "t",
+    "e",
+    "x",
+    "t",
+    "s",
+    "\n",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    "s",
+    "e",
+    "l",
+    "f",
+    ".",
+    "l",
+    "a",
+    "b",
+    "e",
+    "l",
+    "s",
+    " ",
+    "=",
+    " ",
+    "l",
+    "a",
+    "b",
+    "e",
+    "l",
+    "s",
+    "\n",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    "s",
+    "e",
+    "l",
+    "f",
+    ".",
+    "f",
+    "t",
+    "_",
+    "m",
+    "o",
+    "d",
+    "e",
+    "l",
+    " ",
+    "=",
+    " ",
+    "f",
+    "a",
+    "s",
+    "t",
+    "t",
+    "e",
+    "x",
+    "t",
+    "_",
+    "m",
+    "o",
+    "d",
+    "e",
+    "l",
+    "\n",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    "s",
+    "e",
+    "l",
+    "f",
+    ".",
+    "m",
+    "a",
+    "x",
+    "_",
+    "l",
+    "e",
+    "n",
+    " ",
+    "=",
+    " ",
+    "m",
+    "a",
+    "x",
+    "_",
+    "l",
+    "e",
+    "n",
+    "\n",
+    "\n",
+    " ",
+    " ",
+    " ",
+    " ",
+    "d",
+    "e",
+    "f",
+    " ",
+    "_",
+    "_",
+    "l",
+    "e",
+    "n",
+    "_",
+    "_",
+    "(",
+    "s",
+    "e",
+    "l",
+    "f",
+    ")",
+    ":",
+    "\n",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    "r",
+    "e",
+    "t",
+    "u",
+    "r",
+    "n",
+    " ",
+    "l",
+    "e",
+    "n",
+    "(",
+    "s",
+    "e",
+    "l",
+    "f",
+    ".",
+    "t",
+    "e",
+    "x",
+    "t",
+    "s",
+    ")",
+    "\n",
+    "\n",
+    " ",
+    " ",
+    " ",
+    " ",
+    "d",
+    "e",
+    "f",
+    " ",
+    "_",
+    "_",
+    "g",
+    "e",
+    "t",
+    "i",
+    "t",
+    "e",
+    "m",
+    "_",
+    "_",
+    "(",
+    "s",
+    "e",
+    "l",
+    "f",
+    ",",
+    " ",
+    "i",
+    "d",
+    "x",
+    ")",
+    ":",
+    "\n",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    "t",
+    "e",
+    "x",
+    "t",
+    " ",
+    "=",
+    " ",
+    "s",
+    "t",
+    "r",
+    "(",
+    "s",
+    "e",
+    "l",
+    "f",
+    ".",
+    "t",
+    "e",
+    "x",
+    "t",
+    "s",
+    "[",
+    "i",
+    "d",
+    "x",
+    "]",
+    ")",
+    "\n",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    "l",
+    "a",
+    "b",
+    "e",
+    "l",
+    " ",
+    "=",
+    " ",
+    "s",
+    "e",
+    "l",
+    "f",
+    ".",
+    "l",
+    "a",
+    "b",
+    "e",
+    "l",
+    "s",
+    "[",
+    "i",
+    "d",
+    "x",
+    "]",
+    "\n",
+    "\n",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    "#",
+    " ",
+    "V",
+    "e",
+    "c",
+    "t",
+    "o",
+    "r",
+    "i",
+    "z",
+    "e",
+    " ",
+    "o",
+    "n",
+    " ",
+    "t",
+    "h",
+    "e",
+    " ",
+    "f",
+    "l",
+    "y",
+    "\n",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    "w",
+    "o",
+    "r",
+    "d",
+    "s",
+    " ",
+    "=",
+    " ",
+    "t",
+    "e",
+    "x",
+    "t",
+    ".",
+    "s",
+    "p",
+    "l",
+    "i",
+    "t",
+    "(",
+    ")",
+    "[",
+    ":",
+    "s",
+    "e",
+    "l",
+    "f",
+    ".",
+    "m",
+    "a",
+    "x",
+    "_",
+    "l",
+    "e",
+    "n",
+    "]",
+    "\n",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    "v",
+    "e",
+    "c",
+    "s",
+    " ",
+    "=",
+    " ",
+    "[",
+    "s",
+    "e",
+    "l",
+    "f",
+    ".",
+    "f",
+    "t",
+    "_",
+    "m",
+    "o",
+    "d",
+    "e",
+    "l",
+    ".",
+    "g",
+    "e",
+    "t",
+    "_",
+    "w",
+    "o",
+    "r",
+    "d",
+    "_",
+    "v",
+    "e",
+    "c",
+    "t",
+    "o",
+    "r",
+    "(",
+    "w",
+    ")",
+    " ",
+    "f",
+    "o",
+    "r",
+    " ",
+    "w",
+    " ",
+    "i",
+    "n",
+    " ",
+    "w",
+    "o",
+    "r",
+    "d",
+    "s",
+    "]",
+    "\n",
+    "\n",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    "#",
+    " ",
+    "P",
+    "a",
+    "d",
+    "d",
+    "i",
+    "n",
+    "g",
+    "\n",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    "i",
+    "f",
+    " ",
+    "l",
+    "e",
+    "n",
+    "(",
+    "v",
+    "e",
+    "c",
+    "s",
+    ")",
+    " ",
+    "<",
+    " ",
+    "s",
+    "e",
+    "l",
+    "f",
+    ".",
+    "m",
+    "a",
+    "x",
+    "_",
+    "l",
+    "e",
+    "n",
+    ":",
+    "\n",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    "p",
+    "a",
+    "d",
+    "d",
+    "i",
+    "n",
+    "g",
+    " ",
+    "=",
+    " ",
+    "[",
+    "n",
+    "p",
+    ".",
+    "z",
+    "e",
+    "r",
+    "o",
+    "s",
+    "(",
+    "2",
+    "5",
+    "6",
+    ")",
+    " ",
+    "f",
+    "o",
+    "r",
+    " ",
+    "_",
+    " ",
+    "i",
+    "n",
+    " ",
+    "r",
+    "a",
+    "n",
+    "g",
+    "e",
+    "(",
+    "s",
+    "e",
+    "l",
+    "f",
+    ".",
+    "m",
+    "a",
+    "x",
+    "_",
+    "l",
+    "e",
+    "n",
+    " ",
+    "-",
+    " ",
+    "l",
+    "e",
+    "n",
+    "(",
+    "v",
+    "e",
+    "c",
+    "s",
+    ")",
+    ")",
+    "]",
+    "\n",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    "v",
+    "e",
+    "c",
+    "s",
+    ".",
+    "e",
+    "x",
+    "t",
+    "e",
+    "n",
+    "d",
+    "(",
+    "p",
+    "a",
+    "d",
+    "d",
+    "i",
+    "n",
+    "g",
+    ")",
+    "\n",
+    "\n",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    " ",
+    "r",
+    "e",
+    "t",
+    "u",
+    "r",
+    "n",
+    " ",
+    "t",
+    "o",
+    "r",
+    "c",
+    "h",
+    ".",
+    "t",
+    "e",
+    "n",
+    "s",
+    "o",
+    "r",
+    "(",
+    "n",
+    "p",
+    ".",
+    "a",
+    "r",
+    "r",
+    "a",
+    "y",
+    "(",
+    "v",
+    "e",
+    "c",
+    "s",
+    ")",
+    ",",
+    " ",
+    "d",
+    "t",
+    "y",
+    "p",
+    "e",
+    "=",
+    "t",
+    "o",
+    "r",
+    "c",
+    "h",
+    ".",
+    "f",
+    "l",
+    "o",
+    "a",
+    "t",
+    "3",
+    "2",
+    ")",
+    ",",
+    " ",
+    "t",
+    "o",
+    "r",
+    "c",
+    "h",
+    ".",
+    "t",
+    "e",
+    "n",
+    "s",
+    "o",
+    "r",
+    "(",
+    "[",
+    "l",
+    "a",
+    "b",
+    "e",
+    "l",
+    "]",
+    ",",
+    " ",
+    "d",
+    "t",
+    "y",
+    "p",
+    "e",
+    "=",
+    "t",
+    "o",
+    "r",
+    "c",
+    "h",
+    ".",
+    "f",
+    "l",
+    "o",
+    "a",
+    "t",
+    "3",
+    "2",
+    ")",
+    "\n",
+    "\n",
+    "#",
+    " ",
+    "I",
+    "m",
+    "p",
+    "l",
+    "e",
+    "m",
+    "e",
+    "n",
+    "t",
+    "a",
+    "t",
+    "i",
+    "o",
+    "n",
+    "\n",
+    "X",
+    "_",
+    "t",
+    "e",
+    "x",
+    "t",
+    " ",
+    "=",
+    " ",
+    "c",
+    "o",
+    "r",
+    "p",
+    "u",
+    "s",
+    "_",
+    "f",
+    "u",
+    "l",
+    "l",
+    "[",
+    "'",
+    "t",
+    "e",
+    "x",
+    "t",
+    "'",
+    "]",
+    ".",
+    "v",
+    "a",
+    "l",
+    "u",
+    "e",
+    "s",
+    "\n",
+    "y",
+    " ",
+    "=",
+    " ",
+    "(",
+    "c",
+    "o",
+    "r",
+    "p",
+    "u",
+    "s",
+    "_",
+    "f",
+    "u",
+    "l",
+    "l",
+    "[",
+    "'",
+    "l",
+    "a",
+    "b",
+    "e",
+    "l",
+    "'",
+    "]",
+    ".",
+    "v",
+    "a",
+    "l",
+    "u",
+    "e",
+    "s",
+    ".",
+    "a",
+    "s",
+    "t",
+    "y",
+    "p",
+    "e",
+    "(",
+    "f",
+    "l",
+    "o",
+    "a",
+    "t",
+    ")",
+    " ",
+    "+",
+    " ",
+    "1",
+    ")",
+    " ",
+    "/",
+    " ",
+    "2",
+    "\n",
+    "\n",
+    "f",
+    "r",
+    "o",
+    "m",
+    " ",
+    "s",
+    "k",
+    "l",
+    "e",
+    "a",
+    "r",
+    "n",
+    ".",
+    "m",
+    "o",
+    "d",
+    "e",
+    "l",
+    "_",
+    "s",
+    "e",
+    "l",
+    "e",
+    "c",
+    "t",
+    "i",
+    "o",
+    "n",
+    " ",
+    "i",
+    "m",
+    "p",
+    "o",
+    "r",
+    "t",
+    " ",
+    "t",
+    "r",
+    "a",
+    "i",
+    "n",
+    "_",
+    "t",
+    "e",
+    "s",
+    "t",
+    "_",
+    "s",
+    "p",
+    "l",
+    "i",
+    "t",
+    "\n",
+    "t",
+    "r",
+    "a",
+    "i",
+    "n",
+    "_",
+    "t",
+    "x",
+    "t",
+    ",",
+    " ",
+    "v",
+    "a",
+    "l",
+    "_",
+    "t",
+    "x",
+    "t",
+    ",",
+    " ",
+    "t",
+    "r",
+    "a",
+    "i",
+    "n",
+    "_",
+    "y",
+    ",",
+    " ",
+    "v",
+    "a",
+    "l",
+    "_",
+    "y",
+    " ",
+    "=",
+    " ",
+    "t",
+    "r",
+    "a",
+    "i",
+    "n",
+    "_",
+    "t",
+    "e",
+    "s",
+    "t",
+    "_",
+    "s",
+    "p",
+    "l",
+    "i",
+    "t",
+    "(",
+    "X",
+    "_",
+    "t",
+    "e",
+    "x",
+    "t",
+    ",",
+    " ",
+    "y",
+    ",",
+    " ",
+    "t",
+    "e",
+    "s",
+    "t",
+    "_",
+    "s",
+    "i",
+    "z",
+    "e",
+    "=",
+    "0",
+    ".",
+    "2",
+    ",",
+    " ",
+    "r",
+    "a",
+    "n",
+    "d",
+    "o",
+    "m",
+    "_",
+    "s",
+    "t",
+    "a",
+    "t",
+    "e",
+    "=",
+    "4",
+    "2",
+    ")",
+    "\n",
+    "\n",
+    "#",
+    " ",
+    "C",
+    "r",
+    "e",
+    "a",
+    "t",
+    "e",
+    " ",
+    "t",
+    "h",
+    "e",
+    " ",
+    "l",
+    "a",
+    "z",
+    "y",
+    " ",
+    "l",
+    "o",
+    "a",
+    "d",
+    "e",
+    "r",
+    "s",
+    "\n",
+    "t",
+    "r",
+    "a",
+    "i",
+    "n",
+    "_",
+    "d",
+    "s",
+    " ",
+    "=",
+    " ",
+    "T",
+    "u",
+    "n",
+    "i",
+    "s",
+    "i",
+    "a",
+    "n",
+    "D",
+    "a",
+    "t",
+    "a",
+    "s",
+    "e",
+    "t",
+    "(",
+    "t",
+    "r",
+    "a",
+    "i",
+    "n",
+    "_",
+    "t",
+    "x",
+    "t",
+    ",",
+    " ",
+    "t",
+    "r",
+    "a",
+    "i",
+    "n",
+    "_",
+    "y",
+    ",",
+    " ",
+    "m",
+    "o",
+    "d",
+    "e",
+    "l",
+    ")",
+    " ",
+    "#",
+    " ",
+    "'",
+    "m",
+    "o",
+    "d",
+    "e",
+    "l",
+    "'",
+    " ",
+    "i",
+    "s",
+    " ",
+    "y",
+    "o",
+    "u",
+    "r",
+    " ",
+    "F",
+    "a",
+    "s",
+    "t",
+    "T",
+    "e",
+    "x",
+    "t",
+    " ",
+    "o",
+    "b",
+    "j",
+    "e",
+    "c",
+    "t",
+    "\n",
+    "v",
+    "a",
+    "l",
+    "_",
+    "d",
+    "s",
+    " ",
+    "=",
+    " ",
+    "T",
+    "u",
+    "n",
+    "i",
+    "s",
+    "i",
+    "a",
+    "n",
+    "D",
+    "a",
+    "t",
+    "a",
+    "s",
+    "e",
+    "t",
+    "(",
+    "v",
+    "a",
+    "l",
+    "_",
+    "t",
+    "x",
+    "t",
+    ",",
+    " ",
+    "v",
+    "a",
+    "l",
+    "_",
+    "y",
+    ",",
+    " ",
+    "m",
+    "o",
+    "d",
+    "e",
+    "l",
+    ")",
+    "\n",
+    "\n",
+    "t",
+    "r",
+    "a",
+    "i",
+    "n",
+    "_",
+    "l",
+    "o",
+    "a",
+    "d",
+    "e",
+    "r",
+    " ",
+    "=",
+    " ",
+    "D",
+    "a",
+    "t",
+    "a",
+    "L",
+    "o",
+    "a",
+    "d",
+    "e",
+    "r",
+    "(",
+    "t",
+    "r",
+    "a",
+    "i",
+    "n",
+    "_",
+    "d",
+    "s",
+    ",",
+    " ",
+    "b",
+    "a",
+    "t",
+    "c",
+    "h",
+    "_",
+    "s",
+    "i",
+    "z",
+    "e",
+    "=",
+    "3",
+    "2",
+    ",",
+    " ",
+    "s",
+    "h",
+    "u",
+    "f",
+    "f",
+    "l",
+    "e",
+    "=",
+    "T",
+    "r",
+    "u",
+    "e",
+    ",",
+    " ",
+    "n",
+    "u",
+    "m",
+    "_",
+    "w",
+    "o",
+    "r",
+    "k",
+    "e",
+    "r",
+    "s",
+    "=",
+    "2",
+    ")",
+    "\n",
+    "v",
+    "a",
+    "l",
+    "_",
+    "l",
+    "o",
+    "a",
+    "d",
+    "e",
+    "r",
+    " ",
+    "=",
+    " ",
+    "D",
+    "a",
+    "t",
+    "a",
+    "L",
+    "o",
+    "a",
+    "d",
+    "e",
+    "r",
+    "(",
+    "v",
+    "a",
+    "l",
+    "_",
+    "d",
+    "s",
+    ",",
+    " ",
+    "b",
+    "a",
+    "t",
+    "c",
+    "h",
+    "_",
+    "s",
+    "i",
+    "z",
+    "e",
+    "=",
+    "3",
+    "2",
+    ",",
+    " ",
+    "s",
+    "h",
+    "u",
+    "f",
+    "f",
+    "l",
+    "e",
+    "=",
+    "F",
+    "a",
+    "l",
+    "s",
+    "e",
+    ",",
+    " ",
+    "n",
+    "u",
+    "m",
+    "_",
+    "w",
+    "o",
+    "r",
+    "k",
+    "e",
+    "r",
+    "s",
+    "=",
+    "0",
+    ")"
+   ],
+   "execution_count": 12,
+   "outputs": []
+  },
+  {
+   "cell_type": "code",
+   "source": [
+    "class EarlyStopping:\n",
+    "    def __init__(self, patience=5, delta=0):\n",
+    "        self.patience = patience\n",
+    "        self.delta = delta\n",
+    "        self.best_score = None\n",
+    "        self.early_stop = False\n",
+    "        self.counter = 0\n",
+    "        self.best_model_state = None\n",
+    "\n",
+    "    def __call__(self, val_loss, model):\n",
+    "        score = -val_loss\n",
+    "\n",
+    "        if self.best_score is None:\n",
+    "            self.best_score = score\n",
+    "            self.best_model_state = model.state_dict()\n",
+    "        elif score < self.best_score + self.delta:\n",
+    "            self.counter += 1\n",
+    "            if self.counter >= self.patience:\n",
+    "                self.early_stop = True\n",
+    "        else:\n",
+    "            self.best_score = score\n",
+    "            self.best_model_state = model.state_dict()\n",
+    "            self.counter = 0\n",
+    "\n",
+    "    def load_best_model(self, model):\n",
+    "        model.load_state_dict(self.best_model_state)"
+   ],
+   "metadata": {
+    "id": "ChlmWz3nEOTZ"
+   },
+   "execution_count": 7,
+   "outputs": []
+  },
+  {
+   "cell_type": "code",
+   "source": [
+    "def label_smoothed_bce(outputs, targets, smoothing=0.15):\n",
+    "    \"\"\"Softens 0\u21920.075 and 1\u21920.925 \u2014 stronger regularisation than 0.1.\"\"\"\n",
+    "    with torch.no_grad():\n",
+    "        targets = targets * (1 - smoothing) + 0.5 * smoothing\n",
+    "    return nn.BCEWithLogitsLoss()(outputs, targets)\n"
+   ],
+   "metadata": {
+    "id": "pOjRjrd1Fbqc"
+   },
+   "execution_count": 8,
+   "outputs": []
+  },
+  {
+   "cell_type": "code",
+   "source": [
+    "device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
+    "\n",
+    "# v3: hidden_size=48, dropout=0.4\n",
+    "net = Asfour(input_dim=256, hidden_size=48, num_classes=1, dropout=0.4).to(device)\n",
+    "\n",
+    "# Stronger weight decay (0.01 \u2192 0.05) to penalise large weights harder\n",
+    "optimizer = torch.optim.AdamW(net.parameters(), lr=1e-3, weight_decay=0.05)\n",
+    "\n",
+    "NUM_EPOCHS = 30\n",
+    "scheduler = torch.optim.lr_scheduler.OneCycleLR(\n",
+    "    optimizer,\n",
+    "    max_lr=1e-3,\n",
+    "    steps_per_epoch=len(train_loader),\n",
+    "    epochs=NUM_EPOCHS,\n",
+    "    pct_start=0.15,       # slightly longer warmup\n",
+    "    anneal_strategy=\"cos\",\n",
+    ")\n",
+    "\n",
+    "# Patience bumped to 7 \u2014 gives the model more room under strong regularisation\n",
+    "es = EarlyStopping(patience=7, delta=0.002)\n",
+    "\n",
+    "for epoch in range(NUM_EPOCHS):\n",
+    "    net.train()\n",
+    "    train_loss = 0\n",
+    "    for batch_x, batch_y in train_loader:\n",
+    "        batch_x, batch_y = batch_x.to(device), batch_y.to(device)\n",
+    "        optimizer.zero_grad()\n",
+    "        outputs = net(batch_x)\n",
+    "        loss = label_smoothed_bce(outputs, batch_y)\n",
+    "        loss.backward()\n",
+    "        torch.nn.utils.clip_grad_norm_(net.parameters(), max_norm=1.0)\n",
+    "        optimizer.step()\n",
+    "        scheduler.step()   # OneCycleLR steps per batch\n",
+    "        train_loss += loss.item()\n",
+    "\n",
+    "    net.eval()\n",
+    "    val_loss = 0\n",
+    "    with torch.no_grad():\n",
+    "        for batch_x, batch_y in val_loader:\n",
+    "            batch_x, batch_y = batch_x.to(device), batch_y.to(device)\n",
+    "            outputs = net(batch_x)\n",
+    "            v_loss = nn.BCEWithLogitsLoss()(outputs, batch_y)\n",
+    "            val_loss += v_loss.item()\n",
+    "\n",
+    "    avg_train = train_loss / len(train_loader)\n",
+    "    avg_val   = val_loss   / len(val_loader)\n",
+    "\n",
+    "    print(f\"Epoch {epoch+1:02d}/{NUM_EPOCHS}: \"\n",
+    "          f\"Train {avg_train:.4f}  Val {avg_val:.4f}  \"\n",
+    "          f\"Gap {avg_val - avg_train:+.4f}  \"\n",
+    "          f\"LR {optimizer.param_groups[0]['lr']:.2e}\")\n",
+    "\n",
+    "    es(avg_val, net)\n",
+    "    if es.early_stop:\n",
+    "        print(\"Early stopping \u2014 loading best checkpoint\")\n",
+    "        es.load_best_model(net)\n",
+    "        break\n"
+   ],
+   "metadata": {
+    "colab": {
+     "base_uri": "https://localhost:8080/"
+    },
+    "id": "-QykeL9J__rL",
+    "outputId": "19c2f7b1-c256-4aa4-ca12-9e5b96ec52b3"
+   },
+   "execution_count": 9,
+   "outputs": [
+    {
+     "output_type": "stream",
+     "name": "stdout",
+     "text": [
+      "Epoch 01/20: Train 0.5372  Val 0.3916  Gap -0.1456  LR 5.20e-04\n",
+      "Epoch 02/20: Train 0.4560  Val 0.3694  Gap -0.0866  LR 1.00e-03\n",
+      "Epoch 03/20: Train 0.4334  Val 0.3507  Gap -0.0827  LR 9.92e-04\n",
+      "Epoch 04/20: Train 0.4195  Val 0.3435  Gap -0.0760  LR 9.70e-04\n",
+      "Epoch 05/20: Train 0.4082  Val 0.3458  Gap -0.0624  LR 9.33e-04\n",
+      "Epoch 06/20: Train 0.3990  Val 0.3487  Gap -0.0503  LR 8.83e-04\n",
+      "Epoch 07/20: Train 0.3891  Val 0.3431  Gap -0.0460  LR 8.21e-04\n",
+      "Epoch 08/20: Train 0.3800  Val 0.3416  Gap -0.0384  LR 7.50e-04\n",
+      "Epoch 09/20: Train 0.3715  Val 0.3430  Gap -0.0285  LR 6.71e-04\n",
+      "Early stopping \u2014 loading best checkpoint\n"
+     ]
+    }
+   ]
+  },
+  {
+   "cell_type": "code",
+   "source": [
+    "from sklearn.metrics import classification_report, roc_auc_score\n",
+    "\n",
+    "net.eval()\n",
+    "all_preds, all_labels = [], []\n",
+    "\n",
+    "with torch.no_grad():\n",
+    "    for batch_x, batch_y in val_loader:\n",
+    "        batch_x = batch_x.to(device)\n",
+    "        outputs = torch.sigmoid(net(batch_x))\n",
+    "        preds = (outputs.cpu() > 0.5).float()\n",
+    "        all_preds.extend(preds.view(-1).tolist())\n",
+    "        all_labels.extend(batch_y.view(-1).tolist())\n",
+    "\n",
+    "print(classification_report(all_labels, all_preds, target_names=[\"clean\", \"profane\"]))\n",
+    "print(\"AUC:\", roc_auc_score(all_labels, all_preds))"
+   ],
+   "metadata": {
+    "id": "cB08BYR6UA2Q",
+    "colab": {
+     "base_uri": "https://localhost:8080/"
+    },
+    "outputId": "bf4aa0af-0a5b-4840-d53c-192810ff4caa"
+   },
+   "execution_count": 10,
+   "outputs": [
+    {
+     "output_type": "stream",
+     "name": "stdout",
+     "text": [
+      "              precision    recall  f1-score   support\n",
+      "\n",
+      "       clean       0.82      0.85      0.83      4281\n",
+      "     profane       0.88      0.85      0.87      5461\n",
+      "\n",
+      "    accuracy                           0.85      9742\n",
+      "   macro avg       0.85      0.85      0.85      9742\n",
+      "weighted avg       0.85      0.85      0.85      9742\n",
+      "\n",
+      "AUC: 0.8510448534833718\n"
+     ]
+    }
+   ]
+  },
+  {
+   "cell_type": "code",
+   "metadata": {
+    "colab": {
+     "base_uri": "https://localhost:8080/",
+     "height": 228
+    },
+    "id": "31b32c69",
+    "outputId": "5a467d40-7cf9-40ca-8418-2768aa16e430"
+   },
+   "source": [
+    "from sklearn.metrics import accuracy_score\n",
+    "\n",
+    "# Final Evaluation\n",
+    "net.eval()\n",
+    "probs = []\n",
+    "with torch.no_grad():\n",
+    "    for batch_x, _ in val_loader:\n",
+    "        probs.extend(torch.sigmoid(net(batch_x.to(device))).cpu().numpy())\n",
+    "\n",
+    "# Try to find the best threshold to maximize accuracy\n",
+    "thresholds = np.linspace(0.1, 0.9, 81)\n",
+    "accuracies = [accuracy_score(all_val_labels, (np.array(probs) > t).astype(int)) for t in thresholds]\n",
+    "best_t = thresholds[np.argmax(accuracies)]\n",
+    "\n",
+    "print(f\"Best Threshold: {best_t:.2f}\")\n",
+    "print(f\"Final Accuracy at Best Threshold: {max(accuracies):.4f}\")\n",
+    "\n",
+    "# Final Report\n",
+    "final_preds = (np.array(probs) > best_t).astype(int)\n",
+    "print(classification_report(all_val_labels, final_preds, target_names=[\"clean\", \"profane\"]))"
+   ],
+   "execution_count": 11,
+   "outputs": [
+    {
+     "output_type": "error",
+     "ename": "NameError",
+     "evalue": "name 'all_val_labels' is not defined",
+     "traceback": [
+      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
+      "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
+      "\u001b[0;32m/tmp/ipykernel_1236/2364656557.py\u001b[0m in \u001b[0;36m<cell line: 0>\u001b[0;34m()\u001b[0m\n\u001b[1;32m     10\u001b[0m \u001b[0;31m# Try to find the best threshold to maximize accuracy\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     11\u001b[0m \u001b[0mthresholds\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mlinspace\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m0.1\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m0.9\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m81\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 12\u001b[0;31m \u001b[0maccuracies\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0maccuracy_score\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mall_val_labels\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m(\u001b[0m\u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0marray\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mprobs\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;34m>\u001b[0m \u001b[0mt\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mastype\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mint\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;32mfor\u001b[0m \u001b[0mt\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mthresholds\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     13\u001b[0m \u001b[0mbest_t\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mthresholds\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0margmax\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0maccuracies\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     14\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
+      "\u001b[0;31mNameError\u001b[0m: name 'all_val_labels' is not defined"
+     ]
+    }
+   ]
+  },
+  {
+   "cell_type": "code",
+   "source": [
+    "def count_params(model):\n",
+    "    total = sum(p.numel() for p in model.parameters())\n",
+    "    trainable = sum(p.numel() for p in model.parameters() if p.requires_grad)\n",
+    "    print(f\"Total params:     {total:,}\")\n",
+    "    print(f\"Trainable params: {trainable:,}\")"
+   ],
+   "metadata": {
+    "id": "V6X-oV-qXKvu"
+   },
+   "execution_count": null,
+   "outputs": []
+  },
+  {
+   "cell_type": "code",
+   "source": [
+    "count_params(net)"
+   ],
+   "metadata": {
+    "colab": {
+     "base_uri": "https://localhost:8080/"
+    },
+    "id": "12esUm_uXK-W",
+    "outputId": "e845b9f5-0086-4118-d4fb-73fc5dc030a1"
+   },
+   "execution_count": null,
+   "outputs": [
+    {
+     "output_type": "stream",
+     "name": "stdout",
+     "text": [
+      "Total params:     2,433,281\n",
+      "Trainable params: 2,433,281\n"
+     ]
+    }
+   ]
+  },
+  {
+   "cell_type": "code",
+   "source": [
+    "# \u2500\u2500 Inference helper (fixed: uses word vectors, not sentence vector) \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\n",
+    "def predict(text, ft_model, net, device, max_len=30, threshold=0.5):\n",
+    "    \"\"\"Correctly feeds per-word fastText vectors into Asfour.\"\"\"\n",
+    "    words = text.split()[:max_len]\n",
+    "    vecs  = [ft_model.get_word_vector(w) for w in words]\n",
+    "    if len(vecs) < max_len:\n",
+    "        import numpy as np\n",
+    "        vecs += [np.zeros(256)] * (max_len - len(vecs))\n",
+    "    import numpy as np\n",
+    "    x = torch.tensor(np.array(vecs), dtype=torch.float32)  # [T, 256]\n",
+    "    x = x.unsqueeze(0).to(device)                          # [1, T, 256]\n",
+    "    net.eval()\n",
+    "    with torch.no_grad():\n",
+    "        prob = torch.sigmoid(net(x)).item()\n",
+    "    label = 'profane' if prob > threshold else 'clean'\n",
+    "    print(f\"Text : '{text}'\")\n",
+    "    print(f\"Prob : {prob:.3f}  \u2192  {label}\")\n",
+    "    return prob\n",
+    "\n",
+    "\n",
+    "sample_text = \"3asba lik ye bechirMNST\"\n",
+    "predict(sample_text, model, net, device)\n"
+   ],
+   "metadata": {
+    "colab": {
+     "base_uri": "https://localhost:8080/"
+    },
+    "id": "GfY49fuEXTWp",
+    "outputId": "c6617fc4-9377-4b6f-dd86-e748274415d1"
+   },
+   "execution_count": null,
+   "outputs": [
+    {
+     "output_type": "stream",
+     "name": "stdout",
+     "text": [
+      "\n",
+      "Text: '3asba lik ye bechirMNST'\n",
+      "Profanity probability: 0.010\n",
+      "Prediction: profane\n"
+     ]
+    }
+   ]
+  },
+  {
+   "cell_type": "code",
+   "source": [],
+   "metadata": {
+    "id": "7Zvj40YIXWfb"
+   },
+   "execution_count": null,
+   "outputs": []
+  },
+  {
+   "cell_type": "code",
+   "metadata": {
+    "id": "53114288"
+   },
+   "source": [
+    "save_path = '/content/drive/MyDrive/Asfour_V3.pt'\n",
+    "torch.save({\n",
+    "    'model_state_dict': net.state_dict(),\n",
+    "    'optimizer_state_dict': optimizer.state_dict(),\n",
+    "    'best_threshold': best_t,\n",
+    "    'hyperparameters': {\n",
+    "        'input_dim': 256,\n",
+    "        'hidden_size': 48,\n",
+    "        'num_classes': 1,\n",
+    "        'dropout': 0.4,\n",
+    "        'version': 'v3',\n",
+    "    }\n",
+    "}, save_path)\n",
+    "\n",
+    "print(f\"Model saved to: {save_path}\")\n"
+   ],
+   "execution_count": null,
+   "outputs": []
+  }
+ ]
+}