Upload 01_generate_fon_translation_candidates (2).ipynb

01_generate_fon_translation_candidates (2).ipynb

@@ -0,0 +1,176 @@
+{
+  "cells": [
+    {
+      "cell_type": "markdown",
+      "metadata": {},
+      "source": [
+        "# Génération des candidats de traduction Fon–Français\n",
+        "\n",
+        "Ce notebook génère des candidats de traduction automatique pour les scénarios en Fon. Les traductions sont ensuite vérifiées humainement et ne constituent pas les références finales sans validation.\n"
+      ]
+    },
+    {
+      "cell_type": "markdown",
+      "metadata": {},
+      "source": [
+        "## 1. Installation\n"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": null,
+      "metadata": {},
+      "outputs": [],
+      "source": [
+        "!pip install -q transformers sentencepiece accelerate pandas openpyxl\n"
+      ]
+    },
+    {
+      "cell_type": "markdown",
+      "metadata": {},
+      "source": [
+        "## 2. Imports\n"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": null,
+      "metadata": {},
+      "outputs": [],
+      "source": [
+        "import pandas as pd\n",
+        "from transformers import AutoTokenizer, AutoModelForSeq2SeqLM\n",
+        "import torch\n"
+      ]
+    },
+    {
+      "cell_type": "markdown",
+      "metadata": {},
+      "source": [
+        "## 3. Chargement du modèle de traduction\n",
+        "\n",
+        "Le modèle produit uniquement des candidats de traduction ; une vérification humaine est nécessaire.\n"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": null,
+      "metadata": {},
+      "outputs": [],
+      "source": [
+        "MODEL_NAME = \"Helsinki-NLP/opus-mt-fon-fr\"\n",
+        "\n",
+        "tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)\n",
+        "model = AutoModelForSeq2SeqLM.from_pretrained(MODEL_NAME)\n",
+        "device = \"cuda\" if torch.cuda.is_available() else \"cpu\"\n",
+        "model = model.to(device)\n",
+        "print(\"Device:\", device)\n"
+      ]
+    },
+    {
+      "cell_type": "markdown",
+      "metadata": {},
+      "source": [
+        "## 4. Fonction de traduction\n"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": null,
+      "metadata": {},
+      "outputs": [],
+      "source": [
+        "def traduire_fon_vers_fr(texte):\n",
+        "    inputs = tokenizer(texte, return_tensors=\"pt\", truncation=True).to(device)\n",
+        "    with torch.no_grad():\n",
+        "        outputs = model.generate(**inputs, max_new_tokens=128)\n",
+        "    return tokenizer.decode(outputs[0], skip_special_tokens=True)\n"
+      ]
+    },
+    {
+      "cell_type": "markdown",
+      "metadata": {},
+      "source": [
+        "## 5. Chargement des scénarios\n"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": null,
+      "metadata": {},
+      "outputs": [],
+      "source": [
+        "scenarios = pd.read_csv(\"scenarios_40.csv\")\n",
+        "print(\"Nombre de scénarios :\", len(scenarios))\n",
+        "display(scenarios.head())\n"
+      ]
+    },
+    {
+      "cell_type": "markdown",
+      "metadata": {},
+      "source": [
+        "## 6. Génération des traductions candidates\n",
+        "\n",
+        "Adapter le nom de la colonne Fon si nécessaire.\n"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": null,
+      "metadata": {},
+      "outputs": [],
+      "source": [
+        "fon_col = \"fon_text\"\n",
+        "if fon_col not in scenarios.columns:\n",
+        "    raise ValueError(f\"Colonne '{fon_col}' introuvable. Colonnes disponibles : {list(scenarios.columns)}\")\n",
+        "audit = scenarios[[\"scenario_id\", \"category\", fon_col]].copy()\n",
+        "audit[\"traduction_candidate\"] = audit[fon_col].apply(traduire_fon_vers_fr)\n",
+        "display(audit.head(10))\n"
+      ]
+    },
+    {
+      "cell_type": "markdown",
+      "metadata": {},
+      "source": [
+        "## 7. Export pour vérification humaine\n"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": null,
+      "metadata": {},
+      "outputs": [],
+      "source": [
+        "audit[\"traduction_correcte\"] = \"\"\n",
+        "audit[\"fon_corrige\"] = \"\"\n",
+        "audit[\"remarques\"] = \"\"\n",
+        "audit.to_csv(\"translation_audit_40.csv\", index=False, encoding=\"utf-8-sig\")\n",
+        "audit.to_excel(\"translation_audit_40.xlsx\", index=False)\n",
+        "print(\"Fichiers créés : translation_audit_40.csv et translation_audit_40.xlsx\")\n"
+      ]
+    },
+    {
+      "cell_type": "markdown",
+      "metadata": {},
+      "source": [
+        "## 8. Rappel méthodologique\n",
+        "\n",
+        "- Les traductions automatiques sont des candidats.\n",
+        "- Elles doivent être vérifiées humainement avant utilisation comme référence.\n",
+        "- Les 40 traductions Fon du benchmark final ont été validées manuellement.\n"
+      ]
+    }
+  ],
+  "metadata": {
+    "kernelspec": {
+      "display_name": "Python 3",
+      "language": "python",
+      "name": "python3"
+    },
+    "language_info": {
+      "name": "python"
+    }
+  },
+  "nbformat": 4,
+  "nbformat_minor": 5
+}