Copy_of_fast_translator.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "private_outputs": true,
      "provenance": []
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    }
  },
  "cells": [
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "BWAU8rjEypUL"
      },
      "outputs": [],
      "source": [
        "!pip install pysrt requests sentencepiece transformers==4.33 sacremoses  sacrebleu"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "import requests\n",
        "import pysrt\n",
        "from transformers import AutoModelForSeq2SeqLM, AutoTokenizer\n",
        "import torch\n",
        "import re\n",
        "from google.colab import files\n",
        "\n"
      ],
      "metadata": {
        "id": "WeEGPnyvzO20"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "model_name =  \"zaanind/nllb-ensi-v1-tuning\" #\"zaanind/nllb-ensi-v1-tuning-subs\" #\"facebook/nllb-200-distilled-600M\"\n",
        "tokenizer = AutoTokenizer.from_pretrained(model_name)\n",
        "model = AutoModelForSeq2SeqLM.from_pretrained(model_name)\n",
        "def dotranslate(text):\n",
        "  inputs = tokenizer(text, return_tensors=\"pt\")\n",
        "  translated_tokens = model.generate(**inputs, forced_bos_token_id=tokenizer.lang_code_to_id[\"sin_Sinh\"])\n",
        "  out = tokenizer.decode(translated_tokens[0], skip_special_tokens=True)\n",
        "  return out\n"
      ],
      "metadata": {
        "id": "LQ3gnm9A0-w8"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [],
      "metadata": {
        "id": "jS1f7lMBzxwD"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "\n",
        "print(\"<< Upload Subtitle file to translate >>\")\n",
        "uploaded = files.upload()\n",
        "\n",
        "for fn in uploaded.keys():\n",
        "  filenm = fn\n",
        "  print(fn)\n",
        "\n",
        "\n",
        "print(\"<<Translating subtitle in progress>>\")\n",
        "\n",
        "# Read the subtitles from a file using pysrt\n",
        "subtitles = pysrt.open(filenm)\n",
        "\n",
        "# Iterate through each subtitle line and translate\n",
        "for subtitle in subtitles:\n",
        "    input_text = subtitle.text\n",
        "\n",
        "    translated_text = dotranslate(input_text)\n",
        "    print(translated_text)\n",
        "\n",
        "    # Update the subtitle with the translated text\n",
        "    subtitle.text = translated_text\n",
        "\n",
        "\n",
        "\n",
        "\n"
      ],
      "metadata": {
        "id": "jqj2S5Puzw6e"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [],
      "metadata": {
        "id": "2goSldRYtPgo"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# @title Download Translated SRT\n",
        "\n",
        "fname = f\"translated_{filenm}\"\n",
        "# Save the modified subtitles to a new file\n",
        "subtitles.save(fname)\n",
        "\n",
        "print(f\"<<downloading your file: {fname} \\n if not, download from file manager>>\")\n",
        "\n",
        "files.download(fname)"
      ],
      "metadata": {
        "id": "sOfzgyoEtPdu"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [],
      "metadata": {
        "id": "y6DwsCM2KNtZ"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [],
      "metadata": {
        "id": "_3YajSOzS1eF"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [],
      "metadata": {
        "id": "bxGWOaDDhh_G"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [],
      "metadata": {
        "id": "W30IN1OOhh67"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [],
      "metadata": {
        "id": "3KKii-mFS1bM"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# @title batch translator\n",
        "\n",
        "batch_size = 3 #num of lines per time\n",
        "\n",
        "print(\"<< Upload Subtitle file to translate >>\")\n",
        "uploaded = files.upload()\n",
        "\n",
        "for fn in uploaded.keys():\n",
        "  filenm = fn\n",
        " # print(fn)\n",
        "\n",
        "\n",
        "print(\"<<Translating subtitle in progress>>\")\n",
        "\n",
        "\n",
        "# Read the subtitles from a file using pysrt\n",
        "subtitles = pysrt.open(filenm)\n",
        "\n",
        "\n",
        "for subtitle in subtitles:\n",
        "    input_text = subtitle.text\n",
        "    text = f\"<sb>{input_text}</sb>\"\n",
        "    subtitle.text = text\n",
        "\n",
        "\n",
        "\n",
        "\n",
        "for i in range(0, len(subtitles), batch_size):\n",
        "  batch = subtitles[i:i + batch_size]\n",
        "  sendai = \"\"\n",
        "  for item in batch:\n",
        "    sendai = sendai + item.text\n",
        "  outed = dotranslate(sendai)\n",
        "  print(outed)\n",
        "  pattern = r\"<sb>(.*?)</sb>\"\n",
        "  spes = re.findall(pattern, outed)\n",
        "  print(len(spes))\n",
        "  for item,trted in zip(batch,spes):\n",
        "    item.text = trted\n",
        "\n",
        "\n",
        "\n",
        "fname = f\"translated_{filenm}\"\n",
        "# Save the modified subtitles to a new file\n",
        "subtitles.save(fname)\n",
        "\n",
        "print(f\"<<downloading your file: {fname} \\n if not, download from file manager>>\")\n",
        "\n",
        "files.download(fname)\n"
      ],
      "metadata": {
        "id": "SVKxxpwYnHhW",
        "cellView": "form"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [],
      "metadata": {
        "id": "y7k7KXX3lIwM"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [],
      "metadata": {
        "id": "XpzZKiKRkP1f"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [],
      "metadata": {
        "id": "NdUDlIDej9gr"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [],
      "metadata": {
        "id": "QqvdfbLApFP7"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "def translate(text): #define function for generation\n",
        "  inputs = tokenizer(text, return_tensors=\"pt\")\n",
        "  translated_tokens = model.generate(**inputs, forced_bos_token_id=tokenizer.lang_code_to_id[\"sin_Sinh\"])\n",
        "  out = tokenizer.decode(translated_tokens[0], skip_special_tokens=True)\n",
        "  return out\n",
        "\n",
        "\n",
        "translate(\"hello how are you?\") #translate"
      ],
      "metadata": {
        "id": "BDN_IpeqwlM_"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "\n"
      ],
      "metadata": {
        "id": "pQ0xu8cEpF5C"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [],
      "metadata": {
        "id": "XsnoXjrnnHII"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [],
      "metadata": {
        "id": "atB5TgUPqw82"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [],
      "metadata": {
        "id": "Zj5icZbisH3i"
      },
      "execution_count": null,
      "outputs": []
    }
  ]
}