{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "1f175efa",
   "metadata": {},
   "source": [
    "# Latin Interpunctuator\n",
    "\n",
    "This notebook demonstrates how to use the mt5 model `mschonhardt/mt5-latin-punctuator-large`.\n",
    "It applies interpunctuation and text formatting standards to Latin text.\n",
    "\n",
    "## Setup Environment"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "id": "044ae4ef",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Torch version: 2.10.0+cu128\n",
      "Device: cuda\n",
      "Environment ready.\n"
     ]
    }
   ],
   "source": [
    "# Import necessary libraries\n",
    "import torch\n",
    "from transformers import AutoTokenizer, AutoModelForSeq2SeqLM\n",
    "\n",
    "# Model should be used with GPU (cuda) if available for faster inference\n",
    "device = \"cuda\" if torch.cuda.is_available() else \"cpu\"\n",
    "\n",
    "print(f\"Torch version: {torch.__version__}\")\n",
    "print(f\"Device: {device}\")\n",
    "\n",
    "print(\"Environment ready.\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4de2def2",
   "metadata": {},
   "source": [
    "## Load the Model from Hugging Face"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "id": "aa5810a8",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Loading model: mschonhardt/mt5-latin-punctuator-large ...\n",
      "Model loaded successfully!\n"
     ]
    }
   ],
   "source": [
    "# Load the model and tokenizer from Huggingface\n",
    "model_name = \"mschonhardt/mt5-latin-punctuator-large\" \n",
    "print(f\"Loading model: {model_name} ...\")\n",
    "tokenizer = AutoTokenizer.from_pretrained(model_name, use_fast=False)\n",
    "model = AutoModelForSeq2SeqLM.from_pretrained(model_name).to(device)\n",
    "print(\"Model loaded successfully!\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2dd05d72",
   "metadata": {},
   "source": [
    "### Prediction Logic\n",
    "Model was trained on prefix \"punctuate: \". `Num_beams` needs to be adjusted when running into hallucinations or repetitions. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "id": "e858df99",
   "metadata": {},
   "outputs": [],
   "source": [
    "def punctuate(text: str) -> str:\n",
    "    # Best practice: Add prefix 'punctuate: 'and lowercase as per training script\n",
    "    input_text = \"punctuate: \" + text.lower()\n",
    "    \n",
    "    inputs = tokenizer(\n",
    "        input_text,\n",
    "        return_tensors=\"pt\",\n",
    "        truncation=True,\n",
    "        max_length=1024,\n",
    "    ).to(device)\n",
    "\n",
    "    with torch.no_grad():\n",
    "        output_ids = model.generate(\n",
    "            **inputs,\n",
    "            max_length=1024,\n",
    "            # Adjust numbeams if hallucination occurs, but 4 is a good starting point for better punctuation\n",
    "            num_beams=4,\n",
    "            early_stopping=True,\n",
    "        )\n",
    "    return tokenizer.decode(output_ids[0], skip_special_tokens=True)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "id": "52fd09e1",
   "metadata": {},
   "outputs": [],
   "source": [
    "text = \"\"\"\n",
    "Si quis Patrem et Filium et Spiritum Sanctum non confitetur tres personas unius substantiae et virtutis ac potestatis, \n",
    "sicut catholica et apostolica ecclesia docet, sed unam tantum ac solitariam dicit esse personam, \n",
    "ita ut ipse sit Pater qui Filius, ipse etiam sit Paraclitus Spiritus, sicut Sabellius et Priscillianus dixerunt, anathema sit.\"\"\""
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e582b0e4",
   "metadata": {},
   "source": [
    "Model was trained on lower case input to prevent overfitting on capital letters and force learning of linguistic pattern."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "id": "6573900a",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " si quis patrem et filium et spiritum sanctum non confitetur tres personas unius\n",
      "substantiae et virtutis ac potestatis  sicut catholica et apostolica ecclesia\n",
      "docet sed unam tantum ac solitariam dicit esse personam  ita ut ipse sit pater\n",
      "qui filius ipse etiam sit paraclitus spiritus sicut sabellius et priscillianus\n",
      "dixerunt anathema sit\n"
     ]
    }
   ],
   "source": [
    "text_without_punctuation = text.replace(\".\",\"\").replace(\",\",\"\").replace(\";\",\"\").replace(\":\",\"\").replace(\"?\",\"\").replace(\"!\",\"\").replace(\"-\",\"\").replace(\"(\",\"\").replace(\")\",\"\").replace(\"[\",\"\").replace(\"]\",\"\").replace(\"{\",\"\").replace(\"}\",\"\").replace(\"\\\"\",\"\")\n",
    "text_without_punctuation = text_without_punctuation.lower()\n",
    "import textwrap\n",
    "print(textwrap.fill(text_without_punctuation, width=80))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "843757c0",
   "metadata": {},
   "source": [
    "### Run Inference"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "id": "86c7521d",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Model will predict punctuation for the input text as well as appropriate use of capital letters\n",
    "# Note: The model will reflect conventions of material it has seen, which might differ from your expectations.\n",
    "text_with_punctuation = punctuate(text_without_punctuation)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "02ef7e70",
   "metadata": {},
   "source": [
    "As the model does apply conventions it has learned from training data, the models decision might differ from your own conventions and expectations. It has not been designed to prepare a 'perfect' text, but to provide structure to unstrucutred text enabling downstream tasks,"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "id": "1c908d35",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Si quis Patrem et Filium et Spiritum sanctum non confitetur tres personas unius\n",
      "substantiae et virtutis ac potestatis, sicut catholica et apostolica Ecclesia\n",
      "docet, sed unam tantum ac solitariam dicit esse personam, ita ut ipse sit Pater\n",
      "qui Filius, ipse etiam sit Paraclitus Spiritus, sicut Sabellius et Priscillianus\n",
      "dixerunt, anathema sit.\n"
     ]
    }
   ],
   "source": [
    "print(textwrap.fill(text_with_punctuation, width=80))"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "venv-jupyter",
   "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": "3.12.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}