옛한글 생성기.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "provenance": [],
      "gpuType": "A100"
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "source": [
        "# Setup"
      ],
      "metadata": {
        "id": "xNW5xe6H-pjx"
      }
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "ttUGCQLS-SbC",
        "outputId": "bfe9e2dd-c68a-4657-87e6-d8506d6ee8d3"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Drive already mounted at /content/drive; to attempt to forcibly remount, call drive.mount(\"/content/drive\", force_remount=True).\n"
          ]
        }
      ],
      "source": [
        "import os\n",
        "from google.colab import userdata, drive\n",
        "\n",
        "os.environ[\"KAGGLE_USERNAME\"] = userdata.get('KAGGLE_USERNAME')\n",
        "os.environ[\"KAGGLE_KEY\"] = userdata.get('KAGGLE_KEY')\n",
        "drive.mount('/content/drive')\n",
        "\n",
        "# Install Keras 3 last. See https://keras.io/getting_started/ for more details.\n",
        "!pip install -q -U keras-nlp datasets\n",
        "!pip install -q -U keras\n",
        "\n",
        "os.environ[\"KERAS_BACKEND\"] = \"tensorflow\"  # \"jax\" Or \"torch\" or \"tensorflow\".\n",
        "# Avoid memory fragmentation on JAX backend.\n",
        "#os.environ[\"XLA_PYTHON_CLIENT_MEM_FRACTION\"]=\"1.00\"\n",
        "\n",
        "# Training Configurations\n",
        "lora_name = \"honggildong\"\n",
        "lora_rank = 8\n",
        "lr_value = 2e-4"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "# Tokenizer Test"
      ],
      "metadata": {
        "id": "zJFT3V0KA8Po"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "import keras_nlp\n",
        "\n",
        "tokenizer = keras_nlp.models.GemmaTokenizer.from_preset(\"gemma_2b_en\")\n",
        "\n",
        "def detoken(tokens):\n",
        "    print(tokens)\n",
        "    for x in tokens.numpy():\n",
        "        word = tokenizer.detokenize([x])\n",
        "        print(f\"{x} -> {word.numpy().decode('utf-8')}\")\n",
        "\n",
        "detoken(tokenizer(\"The quick brown fox jumped.\"))\n",
        "detoken(tokenizer(\"안녕하세요. 반갑습니다. 오늘은 날씨가 참 좋네요.\"))\n",
        "detoken(tokenizer(\"おはようございます。今日はいい天気ですね。\"))\n",
        "detoken(tokenizer(\"“대범 ᄉᆞᄅᆞᆷ이 복ᄌᆡ지한의 나ᄆᆡ 오륜이 이스니 오륜 즁의 군뷔 읏믐이라. ᄉᆞᄅᆞᆷ되고 오륜을 바리면 ᄉᆞᄅᆞᆷ이 아니라 ᄒᆞᄂᆞ니, 이졔 너ᄂᆞᆫ 지혜와 식견이 범ᄉᆞᄅᆞᆷ두곤 더ᄒᆞ되 이ᄅᆞᆯ 모로니 엇지 ᄋᆡ답지 아니ᄒᆞ리요. 우리 셰ᄃᆡ로 국은을 입어 ᄌᆞᄌᆞ손손히 녹을 바드니 망극ᄒᆞᆫ 마음이 갈츙보국ᄒᆞ더니, 우리의게 밋쳐ᄂᆞᆫ 널노 말무야마 역명을 장ᄎᆞ 어ᄂᆡ 곳듸 밋츨 쥴 모로게 되니 엇지 ᄒᆞᆫ심타 ᄲᅮᆫ이며, 난신과 젹ᄌᆞ 어ᄂᆡ ᄃᆡ의 업스리요마난 우리 문호의셔 날 쥴은 진실노 ᄯᅳᆺᄒᆞ지 못ᄒᆞ엿도다. 네의 죄목을 젼하 진로ᄒᆞ시니 맛당이 극형을 ᄒᆡᆼᄒᆞ실 거시로듸, ᄀᆞ지록 성은이 망극ᄒᆞᄉᆞ 죄를 더ᄒᆞ지 아니ᄒᆞ시고 날을 명ᄒᆞᄉᆞ 너을 ᄌᆞ부라 ᄒᆞ옵시니 망극ᄒᆞᆫ 마음 도로혀 황공ᄒᆞ며, 팔십 노친이 ᄇᆡᆨ슈 모년의 널노 ᄒᆞ야곰 쥬야 우려ᄒᆞ시던 즁의 네 이럿탓 변괴을 지어 죄을 나ᄅᆞ의 어드니 놀ᄂᆡ신 마음이 병이 되여 이졔 눕고 장ᄎᆞ 이지 못ᄒᆞ게 되시니, 부친 만일 널노 인ᄒᆞ야 셰상을 ᄇᆞ리시면 네 ᄉᆞᄅᆞ셔도 녁명을 입고, 쥭어 지ᄒᆞ의 간들 쳔츄만ᄃᆡ예 블츙블효지죄ᄅᆞᆯ 유전ᄒᆞᆯ지라. ᄯᅩᄒᆞᆫ 그 나문 우리 일문이 원통치 아니ᄒᆞᆯ야. 네 엇지 넉넉ᄒᆞᆫ 소견으로 이ᄅᆞᆯ ᄉᆡᆼ각지 못ᄒᆞᄂᆞ냐, 네 이 죄명을 ᄀᆞ지고 셰상의 용납할진ᄃᆡ ᄉᆞᄅᆞᆷ읜 비록 안셔ᄒᆞᄂᆞ 소소ᄒᆞᆫ 쳔벌이 ᄉᆞ졍이 잇스랴. 이졔 맛당히 쳔명을 슌슈ᄒᆞ야 조졍 쳐분을 지달일 ᄲᅮᆫ이니 ᄯᅩ 엇지ᄒᆞ리요. 네 일즉 도라오긔을 ᄇᆞᄅᆡ노라.” ᄒᆞ엿더라.\"))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "bxXbJqT0A99W",
        "outputId": "a6b71668-9af7-472d-f39b-660d90c1e476"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "tf.Tensor([   651   4320   8426  25341  32292 235265], shape=(6,), dtype=int32)\n",
            "651 -> The\n",
            "4320 ->  quick\n",
            "8426 ->  brown\n",
            "25341 ->  fox\n",
            "32292 ->  jumped\n",
            "235265 -> .\n",
            "tf.Tensor(\n",
            "[238179 243415 204551 235265  78821 241772  21743 235265  44245 240703\n",
            " 236648 128856 241215 236361 120816 106509 239156 237526 235265], shape=(19,), dtype=int32)\n",
            "238179 -> 안\n",
            "243415 -> 녕\n",
            "204551 -> 하세요\n",
            "235265 -> .\n",
            "78821 ->  반\n",
            "241772 -> 갑\n",
            "21743 -> 습니다\n",
            "235265 -> .\n",
            "44245 ->  오\n",
            "240703 -> 늘\n",
            "236648 -> 은\n",
            "128856 ->  날\n",
            "241215 -> 씨\n",
            "236361 -> 가\n",
            "120816 ->  참\n",
            "106509 ->  좋\n",
            "239156 -> 네\n",
            "237526 -> 요\n",
            "235265 -> .\n",
            "tf.Tensor([130715 235362  40574  10598 106577  14088 235362], shape=(7,), dtype=int32)\n",
            "130715 -> おはようございます\n",
            "235362 -> 。\n",
            "40574 -> 今日は\n",
            "10598 -> いい\n",
            "106577 -> 天気\n",
            "14088 -> ですね\n",
            "235362 -> 。\n",
            "tf.Tensor(\n",
            "[235366 236800 240696 235248 249341 251340 250451 251340 252379 235832\n",
            " 122447 249745    442    351    378 236183 236511 236137  38585 250866\n",
            "    442    351    378  44245 246862 235832  11464 236354 236405  44245\n",
            " 246862 235248    453    383    346 236137 166760 245108 235248    453\n",
            "    374    360 253978 235832 236950 235265 235248 249341 251340 250451\n",
            " 251340 252379 238070 236464  44245 246862 236392  55393 236432 237722\n",
            " 235248 249341 251340 250451 251340 252379 235832 206682 235248 250105\n",
            " 251340 251018 251340 236405 235269  11464    453    378    365 162056\n",
            " 251018 251340 248847  34805 242166 237807 130886 241253 235832 235248\n",
            " 240696 249341 251340 250451 251340 252379 238928 243206  70231 250105\n",
            " 251340 238070  11464 250451 251340 251181  34459 236375 236405 235248\n",
            " 242513 236183 235248 247807    442    351    378 241305 236183 123184\n",
            " 250105 251340 236432 237526 235265  93241 235248 245199 250832    442\n",
            "    351    378 236375  74116 236648 236392  74209 236770 235248 249745\n",
            " 251340 249745 251340 240162 240162 239055 235248 242514 236392  55393\n",
            " 237135 236405 235248 240547 241073 250105 251340 248847  41645 238036\n",
            " 235832 225430    453    401    370 237036 237522 250105 251340 238589\n",
            " 236405 235269  93241 236137 237458 235248 248516 242391 251018 251340\n",
            " 248847 235248 241518 238368  72163 237908 238305 237354  61169 238068\n",
            " 236392  60318 250867 251340  53355 251018    442    351    378 226822\n",
            "    452    364    401 235248 248516 254555 235248 253029  34459 236375\n",
            " 237458 116508 236405 235248 242513 236183 235248 250105 251340 248847\n",
            " 239308 237986 235248    442    349    395 250552 248847 225022 235269\n",
            " 235248 240198 237502 237233 235248    453    377    402 249745 251340\n",
            "  53355 251018    442    351    378 235248 250832    442    351    378\n",
            " 236137 142995 236354 236432 237526 237354 240198  93241  45980 238033\n",
            " 236137 241827 128856 235248 253029 236648  83453 238146 238368 235248\n",
            "    442    349    392 250000 251884 250105 251340 236183 145971 250105\n",
            " 251340 248899 236840 236039 235265 123722 236137 235248 242903 238743\n",
            " 236392 235248 250498 236345  83453 236375 250105 251340 236569 236405\n",
            " 235248 242694 238272 235832 235248 241073 238867 236392 235248 250105\n",
            "    442    351    378 250318 250105 251340 238146  73523 236569 236375\n",
            "    452    364    401 235269 235248 249507 251340 236183 238779  64301\n",
            " 236648 235832 235248 240547 241073 250105 251340 249341 251340 235248\n",
            " 242903 236791  70231 250105 251340 236183 123184 250105 251340 236569\n",
            " 236464 128856 236392  95165 250105 251340 249341 251340 162056 236392\n",
            " 235248 249745 251340 237092 236950 235248 250105 251340 242782 236569\n",
            " 236405 235248 240547 241073 250105 251340 248847  41645 238036  50316\n",
            " 236375 242975 188184 237495 250105 251340 238986 235269 235248 241634\n",
            " 240274  61138 240471 235832 235248 250367    442    351    378 252288\n",
            " 241601  34459 237029 236137 235248 241518 238368 235248 250105 251340\n",
            " 238305 245212 235248 245705 238305  54087 238994 250105 251340 236569\n",
            " 240080 235248    453    383    346 236137 123722  11464 248333 250416\n",
            "  75980 242453 236392  34805 236770 235248 242903 236392  38585 250451\n",
            " 251340 236137  53355 237135 236405 227719 251018    442    351    378\n",
            " 237502  41645 238036 235832 192986 235832 116508 237386  11464    453\n",
            "    378    365 235248 250780 236464  60318 250867 251340  11464 236183\n",
            " 145971 250105 251340 237458 116508 236569 236405 235269  43761 240471\n",
            "  49427 236666 235248 241518 238368  30743 250105 251340 238305 235248\n",
            " 245199 237047 236392 235248 250367 251340 236432 236569 237722 123722\n",
            " 235248 249341 251340 250451 251340 241827 236840 235248 245029 238068\n",
            " 236392  74209 236464 235269 235248    453    382    390 236770  34805\n",
            " 250105 251340 236137 127149 237290 235248    453    396    365 247447\n",
            " 237598 250832    442    351    378 238748 111276    453    401    370\n",
            " 239299 241078 236183 242903 250451 251340 251181  46355 237045 250105\n",
            " 251340 251181 236183 236950 235265 235248    442    349    392 250681\n",
            " 250105 251340 248847  20350  38585 237465  93241  32929 237465 235832\n",
            "  68586 238693 237924 123184 250105 251340 251181 238305 235265 123722\n",
            " 235248 242513 236183 235248 250442 250442 250105 251340 248847  44997\n",
            " 241253  26291  11464 250451 251340 251181 235248 249341    442    351\n",
            "    378 250318 238316 236183 145971 250105 251340 251018 251340 243383\n",
            " 235269 123722  11464 235248 242903 238068 236392 235248 249507 251340\n",
            " 236183 236464 235248 245199 237047 236137 120340 243017 238080 237589\n",
            " 250832    442    351    378 235248 249341 251340 250451 251340 252379\n",
            "    453    374    373  51732 238779  70685 241827 250105 251340 251018\n",
            " 251340  44997 237433 250105 251340 248847 235248    453    396    365\n",
            " 241588 235832 235248 249341 251340 255773 235832 235248 245419 236354\n",
            " 248738 235265  11464    453    378    365 235248 242694 238272 239055\n",
            " 235248    453    396    365 238068 236392 235248 251871 241601 250105\n",
            " 251340 238305  42916 255773 235248 242391 238304 236392  34805 239522\n",
            " 236666 235248    442    349    395 250552 248847 235832 236405 235248\n",
            "    442    349    392 250681 235248 242513 236183 250105 251340 236432\n",
            " 237526 235265 123722  32929 242579  50316 236950 237410 253434 236392\n",
            " 235248 250367 251340 250451    442    351    378 238368 236950   1816\n",
            " 235248 250105 251340 248899 238589 236950 235265], shape=(747,), dtype=int32)\n",
            "235366 -> “\n",
            "236800 -> 대\n",
            "240696 -> 범\n",
            "235248 ->  \n",
            "249341 -> ᄉ\n",
            "251340 -> ᆞ\n",
            "250451 -> ᄅ\n",
            "251340 -> ᆞ\n",
            "252379 -> ᆷ\n",
            "235832 -> 이\n",
            "122447 ->  복\n",
            "249745 -> ᄌ\n",
            "442 -> �\n",
            "351 -> �\n",
            "378 -> �\n",
            "236183 -> 지\n",
            "236511 -> 한\n",
            "236137 -> 의\n",
            "38585 ->  나\n",
            "250866 -> ᄆ\n",
            "442 -> �\n",
            "351 -> �\n",
            "378 -> �\n",
            "44245 ->  오\n",
            "246862 -> 륜\n",
            "235832 -> 이\n",
            "11464 ->  이\n",
            "236354 -> 스\n",
            "236405 -> 니\n",
            "44245 ->  오\n",
            "246862 -> 륜\n",
            "235248 ->  \n",
            "453 -> �\n",
            "383 -> �\n",
            "346 -> �\n",
            "236137 -> 의\n",
            "166760 ->  군\n",
            "245108 -> 뷔\n",
            "235248 ->  \n",
            "453 -> �\n",
            "374 -> �\n",
            "360 -> �\n",
            "253978 -> 믐\n",
            "235832 -> 이\n",
            "236950 -> 라\n",
            "235265 -> .\n",
            "235248 ->  \n",
            "249341 -> ᄉ\n",
            "251340 -> ᆞ\n",
            "250451 -> ᄅ\n",
            "251340 -> ᆞ\n",
            "252379 -> ᆷ\n",
            "238070 -> 되\n",
            "236464 -> 고\n",
            "44245 ->  오\n",
            "246862 -> 륜\n",
            "236392 -> 을\n",
            "55393 ->  바\n",
            "236432 -> 리\n",
            "237722 -> 면\n",
            "235248 ->  \n",
            "249341 -> ᄉ\n",
            "251340 -> ᆞ\n",
            "250451 -> ᄅ\n",
            "251340 -> ᆞ\n",
            "252379 -> ᆷ\n",
            "235832 -> 이\n",
            "206682 ->  아니라\n",
            "235248 ->  \n",
            "250105 -> ᄒ\n",
            "251340 -> ᆞ\n",
            "251018 -> ᄂ\n",
            "251340 -> ᆞ\n",
            "236405 -> 니\n",
            "235269 -> ,\n",
            "11464 ->  이\n",
            "453 -> �\n",
            "378 -> �\n",
            "365 -> �\n",
            "162056 ->  너\n",
            "251018 -> ᄂ\n",
            "251340 -> ᆞ\n",
            "248847 -> ᆫ\n",
            "34805 ->  지\n",
            "242166 -> 혜\n",
            "237807 -> 와\n",
            "130886 ->  식\n",
            "241253 -> 견\n",
            "235832 -> 이\n",
            "235248 ->  \n",
            "240696 -> 범\n",
            "249341 -> ᄉ\n",
            "251340 -> ᆞ\n",
            "250451 -> ᄅ\n",
            "251340 -> ᆞ\n",
            "252379 -> ᆷ\n",
            "238928 -> 두\n",
            "243206 -> 곤\n",
            "70231 ->  더\n",
            "250105 -> ᄒ\n",
            "251340 -> ᆞ\n",
            "238070 -> 되\n",
            "11464 ->  이\n",
            "250451 -> ᄅ\n",
            "251340 -> ᆞ\n",
            "251181 -> ᆯ\n",
            "34459 ->  모\n",
            "236375 -> 로\n",
            "236405 -> 니\n",
            "235248 ->  \n",
            "242513 -> 엇\n",
            "236183 -> 지\n",
            "235248 ->  \n",
            "247807 -> ᄋ\n",
            "442 -> �\n",
            "351 -> �\n",
            "378 -> �\n",
            "241305 -> 답\n",
            "236183 -> 지\n",
            "123184 ->  아니\n",
            "250105 -> ᄒ\n",
            "251340 -> ᆞ\n",
            "236432 -> 리\n",
            "237526 -> 요\n",
            "235265 -> .\n",
            "93241 ->  우리\n",
            "235248 ->  \n",
            "245199 -> 셰\n",
            "250832 -> ᄃ\n",
            "442 -> �\n",
            "351 -> �\n",
            "378 -> �\n",
            "236375 -> 로\n",
            "74116 ->  국\n",
            "236648 -> 은\n",
            "236392 -> 을\n",
            "74209 ->  입\n",
            "236770 -> 어\n",
            "235248 ->  \n",
            "249745 -> ᄌ\n",
            "251340 -> ᆞ\n",
            "249745 -> ᄌ\n",
            "251340 -> ᆞ\n",
            "240162 -> 손\n",
            "240162 -> 손\n",
            "239055 -> 히\n",
            "235248 ->  \n",
            "242514 -> 녹\n",
            "236392 -> 을\n",
            "55393 ->  바\n",
            "237135 -> 드\n",
            "236405 -> 니\n",
            "235248 ->  \n",
            "240547 -> 망\n",
            "241073 -> 극\n",
            "250105 -> ᄒ\n",
            "251340 -> ᆞ\n",
            "248847 -> ᆫ\n",
            "41645 ->  마\n",
            "238036 -> 음\n",
            "235832 -> 이\n",
            "225430 ->  갈\n",
            "453 -> �\n",
            "401 -> �\n",
            "370 -> �\n",
            "237036 -> 보\n",
            "237522 -> 국\n",
            "250105 -> ᄒ\n",
            "251340 -> ᆞ\n",
            "238589 -> 더\n",
            "236405 -> 니\n",
            "235269 -> ,\n",
            "93241 ->  우리\n",
            "236137 -> 의\n",
            "237458 -> 게\n",
            "235248 ->  \n",
            "248516 -> 밋\n",
            "242391 -> 쳐\n",
            "251018 -> ᄂ\n",
            "251340 -> ᆞ\n",
            "248847 -> ᆫ\n",
            "235248 ->  \n",
            "241518 -> 널\n",
            "238368 -> 노\n",
            "72163 ->  말\n",
            "237908 -> 무\n",
            "238305 -> 야\n",
            "237354 -> 마\n",
            "61169 ->  역\n",
            "238068 -> 명\n",
            "236392 -> 을\n",
            "60318 ->  장\n",
            "250867 -> ᄎ\n",
            "251340 -> ᆞ\n",
            "53355 ->  어\n",
            "251018 -> ᄂ\n",
            "442 -> �\n",
            "351 -> �\n",
            "378 -> �\n",
            "226822 ->  곳\n",
            "452 -> �\n",
            "364 -> �\n",
            "401 -> �\n",
            "235248 ->  \n",
            "248516 -> 밋\n",
            "254555 -> 츨\n",
            "235248 ->  \n",
            "253029 -> 쥴\n",
            "34459 ->  모\n",
            "236375 -> 로\n",
            "237458 -> 게\n",
            "116508 ->  되\n",
            "236405 -> 니\n",
            "235248 ->  \n",
            "242513 -> 엇\n",
            "236183 -> 지\n",
            "235248 ->  \n",
            "250105 -> ᄒ\n",
            "251340 -> ᆞ\n",
            "248847 -> ᆫ\n",
            "239308 -> 심\n",
            "237986 -> 타\n",
            "235248 ->  \n",
            "442 -> �\n",
            "349 -> �\n",
            "395 -> �\n",
            "250552 -> ᅮ\n",
            "248847 -> ᆫ\n",
            "225022 -> 이며\n",
            "235269 -> ,\n",
            "235248 ->  \n",
            "240198 -> 난\n",
            "237502 -> 신\n",
            "237233 -> 과\n",
            "235248 ->  \n",
            "453 -> �\n",
            "377 -> �\n",
            "402 -> �\n",
            "249745 -> ᄌ\n",
            "251340 -> ᆞ\n",
            "53355 ->  어\n",
            "251018 -> ᄂ\n",
            "442 -> �\n",
            "351 -> �\n",
            "378 -> �\n",
            "235248 ->  \n",
            "250832 -> ᄃ\n",
            "442 -> �\n",
            "351 -> �\n",
            "378 -> �\n",
            "236137 -> 의\n",
            "142995 ->  업\n",
            "236354 -> 스\n",
            "236432 -> 리\n",
            "237526 -> 요\n",
            "237354 -> 마\n",
            "240198 -> 난\n",
            "93241 ->  우리\n",
            "45980 ->  문\n",
            "238033 -> 호\n",
            "236137 -> 의\n",
            "241827 -> 셔\n",
            "128856 ->  날\n",
            "235248 ->  \n",
            "253029 -> 쥴\n",
            "236648 -> 은\n",
            "83453 ->  진\n",
            "238146 -> 실\n",
            "238368 -> 노\n",
            "235248 ->  \n",
            "442 -> �\n",
            "349 -> �\n",
            "392 -> �\n",
            "250000 -> ᅳ\n",
            "251884 -> ᆺ\n",
            "250105 -> ᄒ\n",
            "251340 -> ᆞ\n",
            "236183 -> 지\n",
            "145971 ->  못\n",
            "250105 -> ᄒ\n",
            "251340 -> ᆞ\n",
            "248899 -> 엿\n",
            "236840 -> 도\n",
            "236039 -> 다\n",
            "235265 -> .\n",
            "123722 ->  네\n",
            "236137 -> 의\n",
            "235248 ->  \n",
            "242903 -> 죄\n",
            "238743 -> 목\n",
            "236392 -> 을\n",
            "235248 ->  \n",
            "250498 -> 젼\n",
            "236345 -> 하\n",
            "83453 ->  진\n",
            "236375 -> 로\n",
            "250105 -> ᄒ\n",
            "251340 -> ᆞ\n",
            "236569 -> 시\n",
            "236405 -> 니\n",
            "235248 ->  \n",
            "242694 -> 맛\n",
            "238272 -> 당\n",
            "235832 -> 이\n",
            "235248 ->  \n",
            "241073 -> 극\n",
            "238867 -> 형\n",
            "236392 -> 을\n",
            "235248 ->  \n",
            "250105 -> ᄒ\n",
            "442 -> �\n",
            "351 -> �\n",
            "378 -> �\n",
            "250318 -> ᆼ\n",
            "250105 -> ᄒ\n",
            "251340 -> ᆞ\n",
            "238146 -> 실\n",
            "73523 ->  거\n",
            "236569 -> 시\n",
            "236375 -> 로\n",
            "452 -> �\n",
            "364 -> �\n",
            "401 -> �\n",
            "235269 -> ,\n",
            "235248 ->  \n",
            "249507 -> ᄀ\n",
            "251340 -> ᆞ\n",
            "236183 -> 지\n",
            "238779 -> 록\n",
            "64301 ->  성\n",
            "236648 -> 은\n",
            "235832 -> 이\n",
            "235248 ->  \n",
            "240547 -> 망\n",
            "241073 -> 극\n",
            "250105 -> ᄒ\n",
            "251340 -> ᆞ\n",
            "249341 -> ᄉ\n",
            "251340 -> ᆞ\n",
            "235248 ->  \n",
            "242903 -> 죄\n",
            "236791 -> 를\n",
            "70231 ->  더\n",
            "250105 -> ᄒ\n",
            "251340 -> ᆞ\n",
            "236183 -> 지\n",
            "123184 ->  아니\n",
            "250105 -> ᄒ\n",
            "251340 -> ᆞ\n",
            "236569 -> 시\n",
            "236464 -> 고\n",
            "128856 ->  날\n",
            "236392 -> 을\n",
            "95165 ->  명\n",
            "250105 -> ᄒ\n",
            "251340 -> ᆞ\n",
            "249341 -> ᄉ\n",
            "251340 -> ᆞ\n",
            "162056 ->  너\n",
            "236392 -> 을\n",
            "235248 ->  \n",
            "249745 -> ᄌ\n",
            "251340 -> ᆞ\n",
            "237092 -> 부\n",
            "236950 -> 라\n",
            "235248 ->  \n",
            "250105 -> ᄒ\n",
            "251340 -> ᆞ\n",
            "242782 -> 옵\n",
            "236569 -> 시\n",
            "236405 -> 니\n",
            "235248 ->  \n",
            "240547 -> 망\n",
            "241073 -> 극\n",
            "250105 -> ᄒ\n",
            "251340 -> ᆞ\n",
            "248847 -> ᆫ\n",
            "41645 ->  마\n",
            "238036 -> 음\n",
            "50316 ->  도\n",
            "236375 -> 로\n",
            "242975 -> 혀\n",
            "188184 ->  황\n",
            "237495 -> 공\n",
            "250105 -> ᄒ\n",
            "251340 -> ᆞ\n",
            "238986 -> 며\n",
            "235269 -> ,\n",
            "235248 ->  \n",
            "241634 -> 팔\n",
            "240274 -> 십\n",
            "61138 ->  노\n",
            "240471 -> 친\n",
            "235832 -> 이\n",
            "235248 ->  \n",
            "250367 -> ᄇ\n",
            "442 -> �\n",
            "351 -> �\n",
            "378 -> �\n",
            "252288 -> ᆨ\n",
            "241601 -> 슈\n",
            "34459 ->  모\n",
            "237029 -> 년\n",
            "236137 -> 의\n",
            "235248 ->  \n",
            "241518 -> 널\n",
            "238368 -> 노\n",
            "235248 ->  \n",
            "250105 -> ᄒ\n",
            "251340 -> ᆞ\n",
            "238305 -> 야\n",
            "245212 -> 곰\n",
            "235248 ->  \n",
            "245705 -> 쥬\n",
            "238305 -> 야\n",
            "54087 ->  우\n",
            "238994 -> 려\n",
            "250105 -> ᄒ\n",
            "251340 -> ᆞ\n",
            "236569 -> 시\n",
            "240080 -> 던\n",
            "235248 ->  \n",
            "453 -> �\n",
            "383 -> �\n",
            "346 -> �\n",
            "236137 -> 의\n",
            "123722 ->  네\n",
            "11464 ->  이\n",
            "248333 -> 럿\n",
            "250416 -> 탓\n",
            "75980 ->  변\n",
            "242453 -> 괴\n",
            "236392 -> 을\n",
            "34805 ->  지\n",
            "236770 -> 어\n",
            "235248 ->  \n",
            "242903 -> 죄\n",
            "236392 -> 을\n",
            "38585 ->  나\n",
            "250451 -> ᄅ\n",
            "251340 -> ᆞ\n",
            "236137 -> 의\n",
            "53355 ->  어\n",
            "237135 -> 드\n",
            "236405 -> 니\n",
            "227719 ->  놀\n",
            "251018 -> ᄂ\n",
            "442 -> �\n",
            "351 -> �\n",
            "378 -> �\n",
            "237502 -> 신\n",
            "41645 ->  마\n",
            "238036 -> 음\n",
            "235832 -> 이\n",
            "192986 ->  병\n",
            "235832 -> 이\n",
            "116508 ->  되\n",
            "237386 -> 여\n",
            "11464 ->  이\n",
            "453 -> �\n",
            "378 -> �\n",
            "365 -> �\n",
            "235248 ->  \n",
            "250780 -> 눕\n",
            "236464 -> 고\n",
            "60318 ->  장\n",
            "250867 -> ᄎ\n",
            "251340 -> ᆞ\n",
            "11464 ->  이\n",
            "236183 -> 지\n",
            "145971 ->  못\n",
            "250105 -> ᄒ\n",
            "251340 -> ᆞ\n",
            "237458 -> 게\n",
            "116508 ->  되\n",
            "236569 -> 시\n",
            "236405 -> 니\n",
            "235269 -> ,\n",
            "43761 ->  부\n",
            "240471 -> 친\n",
            "49427 ->  만\n",
            "236666 -> 일\n",
            "235248 ->  \n",
            "241518 -> 널\n",
            "238368 -> 노\n",
            "30743 ->  인\n",
            "250105 -> ᄒ\n",
            "251340 -> ᆞ\n",
            "238305 -> 야\n",
            "235248 ->  \n",
            "245199 -> 셰\n",
            "237047 -> 상\n",
            "236392 -> 을\n",
            "235248 ->  \n",
            "250367 -> ᄇ\n",
            "251340 -> ᆞ\n",
            "236432 -> 리\n",
            "236569 -> 시\n",
            "237722 -> 면\n",
            "123722 ->  네\n",
            "235248 ->  \n",
            "249341 -> ᄉ\n",
            "251340 -> ᆞ\n",
            "250451 -> ᄅ\n",
            "251340 -> ᆞ\n",
            "241827 -> 셔\n",
            "236840 -> 도\n",
            "235248 ->  \n",
            "245029 -> 녁\n",
            "238068 -> 명\n",
            "236392 -> 을\n",
            "74209 ->  입\n",
            "236464 -> 고\n",
            "235269 -> ,\n",
            "235248 ->  \n",
            "453 -> �\n",
            "382 -> �\n",
            "390 -> �\n",
            "236770 -> 어\n",
            "34805 ->  지\n",
            "250105 -> ᄒ\n",
            "251340 -> ᆞ\n",
            "236137 -> 의\n",
            "127149 ->  간\n",
            "237290 -> 들\n",
            "235248 ->  \n",
            "453 -> �\n",
            "396 -> �\n",
            "365 -> �\n",
            "247447 -> 츄\n",
            "237598 -> 만\n",
            "250832 -> ᄃ\n",
            "442 -> �\n",
            "351 -> �\n",
            "378 -> �\n",
            "238748 -> 예\n",
            "111276 ->  블\n",
            "453 -> �\n",
            "401 -> �\n",
            "370 -> �\n",
            "239299 -> 블\n",
            "241078 -> 효\n",
            "236183 -> 지\n",
            "242903 -> 죄\n",
            "250451 -> ᄅ\n",
            "251340 -> ᆞ\n",
            "251181 -> ᆯ\n",
            "46355 ->  유\n",
            "237045 -> 전\n",
            "250105 -> ᄒ\n",
            "251340 -> ᆞ\n",
            "251181 -> ᆯ\n",
            "236183 -> 지\n",
            "236950 -> 라\n",
            "235265 -> .\n",
            "235248 ->  \n",
            "442 -> �\n",
            "349 -> �\n",
            "392 -> �\n",
            "250681 -> ᅩ\n",
            "250105 -> ᄒ\n",
            "251340 -> ᆞ\n",
            "248847 -> ᆫ\n",
            "20350 ->  그\n",
            "38585 ->  나\n",
            "237465 -> 문\n",
            "93241 ->  우리\n",
            "32929 ->  일\n",
            "237465 -> 문\n",
            "235832 -> 이\n",
            "68586 ->  원\n",
            "238693 -> 통\n",
            "237924 -> 치\n",
            "123184 ->  아니\n",
            "250105 -> ᄒ\n",
            "251340 -> ᆞ\n",
            "251181 -> ᆯ\n",
            "238305 -> 야\n",
            "235265 -> .\n",
            "123722 ->  네\n",
            "235248 ->  \n",
            "242513 -> 엇\n",
            "236183 -> 지\n",
            "235248 ->  \n",
            "250442 -> 넉\n",
            "250442 -> 넉\n",
            "250105 -> ᄒ\n",
            "251340 -> ᆞ\n",
            "248847 -> ᆫ\n",
            "44997 ->  소\n",
            "241253 -> 견\n",
            "26291 -> 으로\n",
            "11464 ->  이\n",
            "250451 -> ᄅ\n",
            "251340 -> ᆞ\n",
            "251181 -> ᆯ\n",
            "235248 ->  \n",
            "249341 -> ᄉ\n",
            "442 -> �\n",
            "351 -> �\n",
            "378 -> �\n",
            "250318 -> ᆼ\n",
            "238316 -> 각\n",
            "236183 -> 지\n",
            "145971 ->  못\n",
            "250105 -> ᄒ\n",
            "251340 -> ᆞ\n",
            "251018 -> ᄂ\n",
            "251340 -> ᆞ\n",
            "243383 -> 냐\n",
            "235269 -> ,\n",
            "123722 ->  네\n",
            "11464 ->  이\n",
            "235248 ->  \n",
            "242903 -> 죄\n",
            "238068 -> 명\n",
            "236392 -> 을\n",
            "235248 ->  \n",
            "249507 -> ᄀ\n",
            "251340 -> ᆞ\n",
            "236183 -> 지\n",
            "236464 -> 고\n",
            "235248 ->  \n",
            "245199 -> 셰\n",
            "237047 -> 상\n",
            "236137 -> 의\n",
            "120340 ->  용\n",
            "243017 -> 납\n",
            "238080 -> 할\n",
            "237589 -> 진\n",
            "250832 -> ᄃ\n",
            "442 -> �\n",
            "351 -> �\n",
            "378 -> �\n",
            "235248 ->  \n",
            "249341 -> ᄉ\n",
            "251340 -> ᆞ\n",
            "250451 -> ᄅ\n",
            "251340 -> ᆞ\n",
            "252379 -> ᆷ\n",
            "453 -> �\n",
            "374 -> �\n",
            "373 -> �\n",
            "51732 ->  비\n",
            "238779 -> 록\n",
            "70685 ->  안\n",
            "241827 -> 셔\n",
            "250105 -> ᄒ\n",
            "251340 -> ᆞ\n",
            "251018 -> ᄂ\n",
            "251340 -> ᆞ\n",
            "44997 ->  소\n",
            "237433 -> 소\n",
            "250105 -> ᄒ\n",
            "251340 -> ᆞ\n",
            "248847 -> ᆫ\n",
            "235248 ->  \n",
            "453 -> �\n",
            "396 -> �\n",
            "365 -> �\n",
            "241588 -> 벌\n",
            "235832 -> 이\n",
            "235248 ->  \n",
            "249341 -> ᄉ\n",
            "251340 -> ᆞ\n",
            "255773 -> 졍\n",
            "235832 -> 이\n",
            "235248 ->  \n",
            "245419 -> 잇\n",
            "236354 -> 스\n",
            "248738 -> 랴\n",
            "235265 -> .\n",
            "11464 ->  이\n",
            "453 -> �\n",
            "378 -> �\n",
            "365 -> �\n",
            "235248 ->  \n",
            "242694 -> 맛\n",
            "238272 -> 당\n",
            "239055 -> 히\n",
            "235248 ->  \n",
            "453 -> �\n",
            "396 -> �\n",
            "365 -> �\n",
            "238068 -> 명\n",
            "236392 -> 을\n",
            "235248 ->  \n",
            "251871 -> 슌\n",
            "241601 -> 슈\n",
            "250105 -> ᄒ\n",
            "251340 -> ᆞ\n",
            "238305 -> 야\n",
            "42916 ->  조\n",
            "255773 -> 졍\n",
            "235248 ->  \n",
            "242391 -> 쳐\n",
            "238304 -> 분\n",
            "236392 -> 을\n",
            "34805 ->  지\n",
            "239522 -> 달\n",
            "236666 -> 일\n",
            "235248 ->  \n",
            "442 -> �\n",
            "349 -> �\n",
            "395 -> �\n",
            "250552 -> ᅮ\n",
            "248847 -> ᆫ\n",
            "235832 -> 이\n",
            "236405 -> 니\n",
            "235248 ->  \n",
            "442 -> �\n",
            "349 -> �\n",
            "392 -> �\n",
            "250681 -> ᅩ\n",
            "235248 ->  \n",
            "242513 -> 엇\n",
            "236183 -> 지\n",
            "250105 -> ᄒ\n",
            "251340 -> ᆞ\n",
            "236432 -> 리\n",
            "237526 -> 요\n",
            "235265 -> .\n",
            "123722 ->  네\n",
            "32929 ->  일\n",
            "242579 -> 즉\n",
            "50316 ->  도\n",
            "236950 -> 라\n",
            "237410 -> 오\n",
            "253434 -> 긔\n",
            "236392 -> 을\n",
            "235248 ->  \n",
            "250367 -> ᄇ\n",
            "251340 -> ᆞ\n",
            "250451 -> ᄅ\n",
            "442 -> �\n",
            "351 -> �\n",
            "378 -> �\n",
            "238368 -> 노\n",
            "236950 -> 라\n",
            "1816 -> .”\n",
            "235248 ->  \n",
            "250105 -> ᄒ\n",
            "251340 -> ᆞ\n",
            "248899 -> 엿\n",
            "238589 -> 더\n",
            "236950 -> 라\n",
            "235265 -> .\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "# Load Dataset"
      ],
      "metadata": {
        "id": "dJuLO9SF-5sY"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "import keras_nlp\n",
        "tokenizer = keras_nlp.models.GemmaTokenizer.from_preset(\"gemma_2b_en\")\n",
        "\n",
        "from datasets import load_dataset\n",
        "ds = load_dataset(\n",
        "    \"bebechien/HongGildongJeon\",\n",
        "    split=\"train\",\n",
        ")\n",
        "print(ds)\n",
        "data = ds.with_format(\"np\", columns=[\"original\"], output_all_columns=False)\n",
        "train = []\n",
        "max_token_len = 0\n",
        "\n",
        "for x in data:\n",
        "  train.append(x['original'])\n",
        "  length = tokenizer(x['original']).get_shape().as_list()[0]\n",
        "  if length > max_token_len:\n",
        "    max_token_len = length\n",
        "    print(f\"longest toekn {max_token_len} - {x['original']}\")\n",
        "\n",
        "print(train[0])\n",
        "print(train[1])\n",
        "print(train[2])\n",
        "\n",
        "print(max_token_len)\n"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "ATGYi1g6-7cd",
        "outputId": "f6567bc7-3a0e-4e06-911c-e88070be24d2"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Attaching 'tokenizer.json' from model 'keras/gemma/keras/gemma_2b_en/2' to your Colab notebook...\n",
            "Attaching 'tokenizer.json' from model 'keras/gemma/keras/gemma_2b_en/2' to your Colab notebook...\n",
            "Attaching 'assets/tokenizer/vocabulary.spm' from model 'keras/gemma/keras/gemma_2b_en/2' to your Colab notebook...\n",
            "/usr/local/lib/python3.10/dist-packages/huggingface_hub/utils/_token.py:88: UserWarning: \n",
            "The secret `HF_TOKEN` does not exist in your Colab secrets.\n",
            "To authenticate with the Hugging Face Hub, create a token in your settings tab (https://huggingface.co/settings/tokens), set it as secret in your Google Colab and restart your session.\n",
            "You will be able to reuse this secret in all of your notebooks.\n",
            "Please note that authentication is recommended but still optional to access public models or datasets.\n",
            "  warnings.warn(\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Dataset({\n",
            "    features: ['original', 'modern translation'],\n",
            "    num_rows: 447\n",
            "})\n",
            "longest toekn 201 - 됴션국셰둉ᄃᆡ왕즉위십오연의홍희문밧긔ᄒᆞᆫᄌᆡ상이잇스되셩은홍이요명은문이니위인이쳥염강직ᄒᆞ여덩망이거록ᄒᆞ니당셰의영웅이라일직용문의올나벼살이할림의쳐ᄒᆞ엿더니명망이됴졍의읏듬되ᄆᆡ젼하그덕망을승이녀긔ᄉᆞ벼살을도도와이조판셔로좌으졍을ᄒᆞ이시니승상이국은을감동ᄒᆞ야갈츙보국ᄒᆞ니ᄉᆞ방의일이업고도젹이업스ᄆᆡ시화연풍ᄒᆞ여나라이ᄐᆡ평ᄒᆞ더라\n",
            "longest toekn 647 - 일일은승상난간의비겨잠ᄀᆞᆫ조의더니ᄒᆞᆫ풍이긜을인도ᄒᆞ여ᄒᆞᆫ고듸다다르니쳥산은암암ᄒᆞ고녹슈난양양ᄒᆞᆫ듸셰류쳔만ᄀᆞ지녹음이파ᄉᆞᄒᆞ고황금갓ᄐᆞᆫᄭᅬᄭᅩ리난춘흥을희롱ᄒᆞ여냥뉴간의왕ᄂᆡᄒᆞ며긔화요초만발ᄒᆞᆫᄃᆡ쳥학ᄇᆡᆨ학이며비취공작이춘광을ᄌᆞ랑ᄒᆞ거날승상이경물을귀경ᄒᆞ며졈졈드러가니만쟝졀벽은하날의다엇고구뷔구뷔벽계슈난골골이폭포되어오운이어러엿난ᄃᆡ길이ᄭᅳᆫ쳐갈바을모로더니문득쳥용이물결을혜치고머리을드러고함ᄒᆞ니산학이믄허지난듯ᄒᆞ더니그용이입을버리고긔운을토ᄒᆞ여승상의입으로드러뵈거날ᄭᆡ다르니평ᄉᆡᆼᄃᆡ몽이라ᄂᆡ염의혜아리되피련군ᄌᆞ을나희리라ᄒᆞ여즉시ᄂᆡ당의드러ᄀᆞ시비을믈이치고부인을익그러취침코져ᄒᆞ니부인이졍ᄉᆡᆨ왈승상은국지ᄌᆡ상이라쳬위존즁ᄒᆞ시거날ᄇᆡᆨ쥬의졍실의드러와노류장화갓치ᄒᆞ시니ᄌᆡ상의쳬면이어ᄃᆡ잇난잇ᄀᆞ승상이ᄉᆡᆼ각ᄒᆞ신직말ᄉᆞᆷ은당연ᄒᆞ오나ᄃᆡ몽을허송할가ᄒᆞ야몽ᄉᆞ을이르지아니ᄒᆞ지고연ᄒᆞ여간쳥ᄒᆞ시니부인이옷슬ᄯᅥᆯ치고밧그로나가시니승상이무류ᄒᆞ신즁의부인의도도ᄒᆞᆫ고집을ᄋᆡ달나무슈히ᄎᆞ탄ᄒᆞ시고외당으로나오시니마ᄎᆞᆷ시비춘셤이상을드리거날좌우고요ᄒᆞ믈인ᄒᆞ여춘셤을잇글고원앙지낙을일의시니져긔울화을더르시나심ᄂᆡ의못ᄂᆡ한탄ᄒᆞ시더라\n",
            "longest toekn 675 - 그방셔의ᄒᆞ여시되대범ᄉᆞᄅᆞᆷ이복ᄌᆡ지한의나ᄆᆡ오륜이이스니오륜즁의군뷔읏믐이라ᄉᆞᄅᆞᆷ되고오륜을바리면ᄉᆞᄅᆞᆷ이아니라ᄒᆞᄂᆞ니이졔너ᄂᆞᆫ지혜와식견이범ᄉᆞᄅᆞᆷ두곤더ᄒᆞ되이ᄅᆞᆯ모로니엇지ᄋᆡ답지아니ᄒᆞ리요우리셰ᄃᆡ로국은을입어ᄌᆞᄌᆞ손손히녹을바드니망극ᄒᆞᆫ마음이갈츙보국ᄒᆞ더니우리의게밋쳐ᄂᆞᆫ널노말무야마역명을장ᄎᆞ어ᄂᆡ곳듸밋츨쥴모로게되니엇지ᄒᆞᆫ심타ᄲᅮᆫ이며난신과젹ᄌᆞ어ᄂᆡᄃᆡ의업스리요마난우리문호의셔날쥴은진실노ᄯᅳᆺᄒᆞ지못ᄒᆞ엿도다네의죄목을젼하진로ᄒᆞ시니맛당이극형을ᄒᆡᆼᄒᆞ실거시로듸ᄀᆞ지록성은이망극ᄒᆞᄉᆞ죄를더ᄒᆞ지아니ᄒᆞ시고날을명ᄒᆞᄉᆞ너을ᄌᆞ부라ᄒᆞ옵시니망극ᄒᆞᆫ마음도로혀황공ᄒᆞ며팔십노친이ᄇᆡᆨ슈모년의널노ᄒᆞ야곰쥬야우려ᄒᆞ시던즁의네이럿탓변괴을지어죄을나ᄅᆞ의어드니놀ᄂᆡ신마음이병이되여이졔눕고장ᄎᆞ이지못ᄒᆞ게되시니부친만일널노인ᄒᆞ야셰상을ᄇᆞ리시면네ᄉᆞᄅᆞ셔도녁명을입고쥭어지ᄒᆞ의간들쳔츄만ᄃᆡ예블츙블효지죄ᄅᆞᆯ유전ᄒᆞᆯ지라ᄯᅩᄒᆞᆫ그나문우리일문이원통치아니ᄒᆞᆯ야네엇지넉넉ᄒᆞᆫ소견으로이ᄅᆞᆯᄉᆡᆼ각지못ᄒᆞᄂᆞ냐네이죄명을ᄀᆞ지고셰상의용납할진ᄃᆡᄉᆞᄅᆞᆷ읜비록안셔ᄒᆞᄂᆞ소소ᄒᆞᆫ쳔벌이ᄉᆞ졍이잇스랴이졔맛당히쳔명을슌슈ᄒᆞ야조졍쳐분을지달일ᄲᅮᆫ이니ᄯᅩ엇지ᄒᆞ리요네일즉도라오긔을ᄇᆞᄅᆡ노라ᄒᆞ엿더라\n",
            "longest toekn 687 - 길동이ᄯᅩᄒᆞᆫ신쟝을분부왈ᄂᆡ몸이조졍의쳐ᄒᆞ여법을ᄌᆞᄇᆞ시면몬쳠불법을업셰여각도ᄉᆞ찰을훼ᄑᆡᄒᆞ렷더니이졔오ᄅᆡ지아니ᄒᆞ여조션국을ᄯᅥ날지라그러ᄒᆞ나부모국이라만리타국의잇셔도잇지못ᄒᆞᆯ지라이졔로각ᄉᆞ의ᄀᆞ혹셰무민ᄒᆞᄂᆞᆫ즁놈을일졔이ᄌᆞᄇᆞ오고ᄯᅩᄒᆞᆫ장안ᄌᆡ상ᄀᆞ의ᄌᆞ식이셰ᄅᆞᆯᄭᅵ고고잔ᄒᆞᆫᄇᆡᆨ셩을소겨ᄌᆡ믈을취ᄒᆞ고블의ᄒᆞᆫ일이만ᄒᆞ며마음이교만ᄒᆞ되구즁이집퍼쳔일이복분의빗초오지못ᄒᆞ고간신이나라의좀미되여셩상의총명을ᄀᆞ리우니ᄀᆞ히ᄒᆞᆫ심ᄒᆞᆫ일이허다ᄒᆞᆫ지라장안의호당지도을낫낫치ᄌᆞᄇᆞ드리라ᄒᆞ니신장이명을듯고공즁으로나라ᄀᆞ더니이시ᄒᆞᆫ후의즁놈ᄇᆡᆨ여명과경화ᄌᆞ졔십여인을ᄌᆞᄇᆞ드리ᄂᆞᆫ지라길동이위염을베플고호령을놉펴각각슈죄왈너희ᄂᆞᆫ다시셰상을보지못ᄒᆞ게ᄒᆞᆯ터이로듸ᄂᆡ몸이나라의조명을ᄇᆞ다국법을ᄌᆞ분ᄇᆡ아니긔로고위안셔ᄒᆞ거니와일후의만일고치지아니ᄒᆞ면너희비록슈만리밧긔잇셔도ᄌᆞᄇᆞ다가버히리라ᄒᆞ고엄형일ᄎᆞ의진문밧긔ᄂᆡ치니라길동이우양을ᄌᆞᄇᆞ군ᄉᆞ을호궤ᄒᆞ고진용을졍졔ᄒᆞ야훤화을금단ᄒᆞ니창쳔만리의ᄇᆡᆨ일이고욕ᄒᆞ고팔진풍운의호령이엄슉ᄒᆞᆫ지라길동이슐을나소와반취ᄒᆞᆫ후의칼을ᄌᆞᄇᆞ츔을츄니검광이분분ᄒᆞ야ᄒᆡ빗츨희롱ᄒᆞ고무슈ᄂᆞᆫ표표ᄒᆞ야공즁의날니ᄂᆞᆫ지라일지셕의라진셰을ᄑᆞᄒᆞ야신장을각각도라보ᄂᆡ고몸을날여활빈당쳐소로도라오니라\n",
            "longest toekn 725 - ᄎᆞ셜길동이그형을이별후의졔군을권ᄒᆞ야농업을심쓰고군법을일ᄉᆞ무며그러구러ᄉᆞᆷ년초토을지ᄂᆡᄆᆡ양식이넉넉ᄒᆞ고슈만군졸이무예와긔보ᄒᆞᄂᆞᆫ법이쳔ᄒᆞ의최ᄀᆞᆼᄒᆞ더라근쳐의ᄒᆞᆫ나라이잇스니일홈은율도국이라즁국을셤긔지아니ᄒᆞ고슈십ᄃᆡᄅᆞᆯ젼ᄌᆞ젼손ᄒᆞ야덕화유ᄒᆡᆼᄒᆞ니나라이ᄐᆡ평ᄒᆞ고ᄇᆡᆨ셩이넉넉ᄒᆞ야날길동이졔군과의논왈우리엇지이도즁만직키여셰월을보ᄂᆡ리요이졔율도국을치고져ᄒᆞ나니각각소견의엇더ᄒᆞ뇨졔인이즐겨원치아니ᄒᆞ리업난지라즉시ᄐᆡᆨ일츌ᄉᆞ할ᄉᆡ삼호걸노션봉을ᄉᆞᆷ고김인슈로후군장을ᄉᆞᆷ고길동스ᄉᆞ로ᄃᆡ원슈되야중영을총독ᄒᆞ니긔병이오쳔이요보졸이이만이라금고ᄒᆞᆷ셩은강산이진동ᄒᆞ고긔치검극은일월을ᄀᆞ리왓더라군ᄉᆞ을ᄌᆡ촉ᄒᆞ여율도국으로향ᄒᆞ니이른바당ᄒᆞᆯᄌᆡ업셔단ᄉᆞ호장으로문을여러항복ᄒᆞᄂᆞᆫ지라슈월지간의칠십여셩을졍ᄒᆞ니위염이일국의진동ᄒᆞᄂᆞᆫ지라도셩오십니밧긔진을치고율도왕의게격셔ᄅᆞᆯ젼ᄒᆞ니그글의ᄒᆞ엿시되의병장홍길동은ᄉᆞᆷᄀᆞ글월을율도왕좌하의드리나니나라난ᄒᆞᆫᄉᆞᄅᆞᆷ이오ᄅᆡ직키지못ᄒᆞᄂᆞᆫ지라시고로셩탕은하걸을치고무왕은상쥬을ᄂᆡ치시니다ᄇᆡᆨ셩을위ᄒᆞ야난ᄃᆡ을평졍ᄒᆞ는ᄇᆡ라이졔의병이십만을거ᄂᆞ러칠십여셩을항복밧고이예이르럿시니왕은ᄃᆡ셰을당ᄒᆞᆯ듯ᄒᆞ거든ᄌᆞ웅을결단ᄒᆞ고셰궁ᄒᆞ거든일직항복ᄒᆞ여쳔명을순슈ᄒᆞ라ᄒᆞ고다시위로왈ᄇᆡᆨ셩을위ᄒᆞ야슈히항셔을올이면일방봉작으로ᄉᆞ직을망케아니ᄒᆞ리라ᄒᆞ여더라\n",
            "longest toekn 770 - 이젹의길동이형지을슈탐ᄒᆞᆫ후의졔장과의논왈명일오시면율도왕을ᄉᆡ로자블거시니군령을어긔오지말나ᄒᆞ고졔장을분발ᄒᆞᆯᄉᆡᄉᆞᆷ호걸을블너왈그ᄃᆡᄂᆞᆫ군ᄉᆞ오쳔을거ᄂᆞ려냥관남편의복병ᄒᆞ엿다ᄀᆞ호령을지다려이리이리ᄒᆞ라ᄒᆞ고후군장김인슈을블너왈그ᄃᆡᄂᆞᆫ군ᄉᆞ이만을거ᄂᆞ려냥관우편의ᄆᆡ복ᄒᆞ엿ᄃᆞᄀᆞ호령을지다려이리이리ᄒᆞ라ᄒᆞ고ᄯᅩ좌션봉ᄆᆡᆼ츙을블너왈그ᄃᆡᄂᆞᆫ쳘긔오쳔을거ᄂᆞ려율왕과ᄊᆞ호다ᄀᆞ거즛ᄑᆡᄒᆞ야왕을인도ᄒᆞ여양관으로다라나다ᄀᆞ츄병양관어귀예들거든이리이리ᄒᆞ라ᄒᆞ고ᄃᆡ장긔치와ᄇᆡᆨ모황월을쥬니라잇틋날평명의ᄆᆡᆼ춘이진문을크게열고ᄃᆡ장긔치을진젼의셰우고웨여왈무도ᄒᆞᆫ율도왕이감히쳔명을항거ᄒᆞ니날을당젹ᄒᆞᆯᄌᆡ조잇거든ᄲᆞᆯ니나와ᄌᆞ웅을결단ᄒᆞ라ᄒᆞ며진문의치돌ᄒᆞ며ᄌᆡ조을비양ᄒᆞ니젹진션봉한셕이웅셩출마왈너희난엇더ᄒᆞᆫ도젹으로쳔위을모로고ᄐᆡ평시졀을블난케ᄒᆞᄂᆞᆫ다오날날너희을ᄉᆡ로잡아민심을안돈ᄒᆞ리라ᄒᆞ고언필의양장이합젼ᄒᆞ야ᄊᆞ호더니슈합이못ᄒᆞ야ᄆᆡᆼ춘의칼이빗나며ᄒᆞᆫ셕의머리을버혀들고좌츙우돌ᄒᆞ여왈율왕은무죄ᄒᆞᆫ장졸을상치말고슈히나황복ᄒᆞ여잔명을보젼ᄒᆞ라ᄒᆞ니율왕이션봉ᄑᆡᄒᆞ믈보고분긔을이긔지못ᄒᆞ야녹포운갑의ᄌᆞ금투고을쓰고좌슈의방쳔극을들고졀니ᄃᆡ완마을ᄌᆡ촉ᄒᆞ야진젼의나셔며왈젹장은잔말말고ᄂᆡ의창을바드라ᄒᆞ고급피ᄆᆡᆼ춘을취ᄒᆞ여ᄊᆞ호니십여합의ᄆᆡᆼ춘이ᄑᆡᄒᆞ여말머리을들너양관으로향ᄒᆞ니율도왕이ᄭᅮ지져왈젹쟝은닷지말고말게나려항복ᄒᆞ라\n",
            "됴션국셰둉ᄃᆡ왕즉위십오연의홍희문밧긔ᄒᆞᆫᄌᆡ상이잇스되셩은홍이요명은문이니위인이쳥염강직ᄒᆞ여덩망이거록ᄒᆞ니당셰의영웅이라일직용문의올나벼살이할림의쳐ᄒᆞ엿더니명망이됴졍의읏듬되ᄆᆡ젼하그덕망을승이녀긔ᄉᆞ벼살을도도와이조판셔로좌으졍을ᄒᆞ이시니승상이국은을감동ᄒᆞ야갈츙보국ᄒᆞ니ᄉᆞ방의일이업고도젹이업스ᄆᆡ시화연풍ᄒᆞ여나라이ᄐᆡ평ᄒᆞ더라\n",
            "일일은승상난간의비겨잠ᄀᆞᆫ조의더니ᄒᆞᆫ풍이긜을인도ᄒᆞ여ᄒᆞᆫ고듸다다르니쳥산은암암ᄒᆞ고녹슈난양양ᄒᆞᆫ듸셰류쳔만ᄀᆞ지녹음이파ᄉᆞᄒᆞ고황금갓ᄐᆞᆫᄭᅬᄭᅩ리난춘흥을희롱ᄒᆞ여냥뉴간의왕ᄂᆡᄒᆞ며긔화요초만발ᄒᆞᆫᄃᆡ쳥학ᄇᆡᆨ학이며비취공작이춘광을ᄌᆞ랑ᄒᆞ거날승상이경물을귀경ᄒᆞ며졈졈드러가니만쟝졀벽은하날의다엇고구뷔구뷔벽계슈난골골이폭포되어오운이어러엿난ᄃᆡ길이ᄭᅳᆫ쳐갈바을모로더니문득쳥용이물결을혜치고머리을드러고함ᄒᆞ니산학이믄허지난듯ᄒᆞ더니그용이입을버리고긔운을토ᄒᆞ여승상의입으로드러뵈거날ᄭᆡ다르니평ᄉᆡᆼᄃᆡ몽이라ᄂᆡ염의혜아리되피련군ᄌᆞ을나희리라ᄒᆞ여즉시ᄂᆡ당의드러ᄀᆞ시비을믈이치고부인을익그러취침코져ᄒᆞ니부인이졍ᄉᆡᆨ왈승상은국지ᄌᆡ상이라쳬위존즁ᄒᆞ시거날ᄇᆡᆨ쥬의졍실의드러와노류장화갓치ᄒᆞ시니ᄌᆡ상의쳬면이어ᄃᆡ잇난잇ᄀᆞ승상이ᄉᆡᆼ각ᄒᆞ신직말ᄉᆞᆷ은당연ᄒᆞ오나ᄃᆡ몽을허송할가ᄒᆞ야몽ᄉᆞ을이르지아니ᄒᆞ지고연ᄒᆞ여간쳥ᄒᆞ시니부인이옷슬ᄯᅥᆯ치고밧그로나가시니승상이무류ᄒᆞ신즁의부인의도도ᄒᆞᆫ고집을ᄋᆡ달나무슈히ᄎᆞ탄ᄒᆞ시고외당으로나오시니마ᄎᆞᆷ시비춘셤이상을드리거날좌우고요ᄒᆞ믈인ᄒᆞ여춘셤을잇글고원앙지낙을일의시니져긔울화을더르시나심ᄂᆡ의못ᄂᆡ한탄ᄒᆞ시더라\n",
            "춘셤이비록쳔인이나ᄌᆡ덕이순직ᄒᆞᆫ지라불의예승상으위엄으로친근ᄒᆞ시니감이위령치못ᄒᆞ여순종ᄒᆞᆫ후로난그날븟텀즁문밧긔나지아니ᄒᆞ고ᄒᆡᆼ실을닥그니그달봇텀ᄐᆡ긔잇셔십ᄉᆡᆨ이당ᄒᆞᄆᆡ거쳐ᄒᆞ는방의오ᄉᆡᆨ운무영농ᄒᆞ며향ᄂᆡ긔히ᄒᆞ더니혼미즁의ᄒᆡᄐᆡᄒᆞ니일ᄀᆡ긔남ᄌᆞ라ᄉᆞᆷ일후의승상이드러와보시니일변긧거오나그쳔ᄉᆡᆼ되믈앗긔시더라일홈을길동이라ᄒᆞ니라이아희졈졈ᄌᆞ라ᄆᆡ긔골이비상ᄒᆞ여ᄒᆞᆫ말을드르면열말을알고ᄒᆞᆫ번보면모로거시업더라일일은승상이길동을다리고ᄂᆡ당의드러ᄀᆞ부인을ᄃᆡᄒᆞ야탄식왈이아히비록영웅이오나쳔ᄉᆡᆼ이라무엇싀쓰리요원통할ᄉᆞ부인의고집이여후회맛급이로소이다부인이그연고을믓ᄌᆞ오니승상이양미을빈츅ᄒᆞ여왈부인이젼일의ᄂᆡ말을드르시던들이아히부인복즁의낫슬낫다엇지쳔ᄉᆡᆼ이되리요인ᄒᆞ여몽ᄉᆞ얼셜화ᄒᆞ시니부인이츄연왈ᄎᆞ역쳔슈오니엇지일력으로ᄒᆞ오릿ᄀᆞ\n",
            "770\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "# Load Model"
      ],
      "metadata": {
        "id": "XhZm69AoAFlT"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "import keras\n",
        "import keras_nlp\n",
        "\n",
        "import time\n",
        "\n",
        "gemma = keras_nlp.models.GemmaCausalLM.from_preset(\"gemma_2b_en\")\n",
        "gemma.summary()\n",
        "\n",
        "tick_start = 0\n",
        "\n",
        "def tick():\n",
        "    global tick_start\n",
        "    tick_start = time.time()\n",
        "\n",
        "def tock():\n",
        "    print(f\"TOTAL TIME ELAPSED: {time.time() - tick_start:.2f}s\")\n",
        "\n",
        "def text_gen(prompt):\n",
        "    tick()\n",
        "    output = gemma.generate(prompt, max_length=256)\n",
        "    print(\"\\nGemma output:\")\n",
        "    print(output)\n",
        "    tock()\n",
        "\n",
        "text_gen(\"나랏말ᄊᆞ미\")\n",
        "text_gen(\"나는월궁의션녀로셔\")\n",
        "text_gen(\"일쟝춘몽이라\")\n",
        "text_gen(\"션녀왈\")\n"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "id": "HynvFo3HAG_O",
        "outputId": "d4478bb3-4536-4f09-b54b-db0b60afe250"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Attaching 'config.json' from model 'keras/gemma/keras/gemma_2b_en/2' to your Colab notebook...\n",
            "Attaching 'config.json' from model 'keras/gemma/keras/gemma_2b_en/2' to your Colab notebook...\n",
            "Attaching 'model.weights.h5' from model 'keras/gemma/keras/gemma_2b_en/2' to your Colab notebook...\n",
            "Attaching 'tokenizer.json' from model 'keras/gemma/keras/gemma_2b_en/2' to your Colab notebook...\n",
            "Attaching 'assets/tokenizer/vocabulary.spm' from model 'keras/gemma/keras/gemma_2b_en/2' to your Colab notebook...\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "\u001b[1mPreprocessor: \"gemma_causal_lm_preprocessor\"\u001b[0m\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\">Preprocessor: \"gemma_causal_lm_preprocessor\"</span>\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓\n",
              "┃\u001b[1m \u001b[0m\u001b[1mTokenizer (type)                                  \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m                                            Vocab #\u001b[0m\u001b[1m \u001b[0m┃\n",
              "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┩\n",
              "│ gemma_tokenizer (\u001b[38;5;33mGemmaTokenizer\u001b[0m)                   │                                             \u001b[38;5;34m256,000\u001b[0m │\n",
              "└────────────────────────────────────────────────────┴─────────────────────────────────────────────────────┘\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓\n",
              "┃<span style=\"font-weight: bold\"> Tokenizer (type)                                   </span>┃<span style=\"font-weight: bold\">                                             Vocab # </span>┃\n",
              "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┩\n",
              "│ gemma_tokenizer (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">GemmaTokenizer</span>)                   │                                             <span style=\"color: #00af00; text-decoration-color: #00af00\">256,000</span> │\n",
              "└────────────────────────────────────────────────────┴─────────────────────────────────────────────────────┘\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "\u001b[1mModel: \"gemma_causal_lm\"\u001b[0m\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\">Model: \"gemma_causal_lm\"</span>\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓\n",
              "┃\u001b[1m \u001b[0m\u001b[1mLayer (type)                 \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1mOutput Shape             \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m        Param #\u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1mConnected to              \u001b[0m\u001b[1m \u001b[0m┃\n",
              "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━┩\n",
              "│ padding_mask (\u001b[38;5;33mInputLayer\u001b[0m)     │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;45mNone\u001b[0m)              │               \u001b[38;5;34m0\u001b[0m │ -                          │\n",
              "├───────────────────────────────┼───────────────────────────┼─────────────────┼────────────────────────────┤\n",
              "│ token_ids (\u001b[38;5;33mInputLayer\u001b[0m)        │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;45mNone\u001b[0m)              │               \u001b[38;5;34m0\u001b[0m │ -                          │\n",
              "├───────────────────────────────┼───────────────────────────┼─────────────────┼────────────────────────────┤\n",
              "│ gemma_backbone                │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m2048\u001b[0m)        │   \u001b[38;5;34m2,506,172,416\u001b[0m │ padding_mask[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m0\u001b[0m],        │\n",
              "│ (\u001b[38;5;33mGemmaBackbone\u001b[0m)               │                           │                 │ token_ids[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m0\u001b[0m]            │\n",
              "├───────────────────────────────┼───────────────────────────┼─────────────────┼────────────────────────────┤\n",
              "│ token_embedding               │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m256000\u001b[0m)      │     \u001b[38;5;34m524,288,000\u001b[0m │ gemma_backbone[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m0\u001b[0m]       │\n",
              "│ (\u001b[38;5;33mReversibleEmbedding\u001b[0m)         │                           │                 │                            │\n",
              "└───────────────────────────────┴───────────────────────────┴─────────────────┴────────────────────────────┘\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓\n",
              "┃<span style=\"font-weight: bold\"> Layer (type)                  </span>┃<span style=\"font-weight: bold\"> Output Shape              </span>┃<span style=\"font-weight: bold\">         Param # </span>┃<span style=\"font-weight: bold\"> Connected to               </span>┃\n",
              "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━┩\n",
              "│ padding_mask (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">InputLayer</span>)     │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>)              │               <span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> │ -                          │\n",
              "├───────────────────────────────┼───────────────────────────┼─────────────────┼────────────────────────────┤\n",
              "│ token_ids (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">InputLayer</span>)        │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>)              │               <span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> │ -                          │\n",
              "├───────────────────────────────┼───────────────────────────┼─────────────────┼────────────────────────────┤\n",
              "│ gemma_backbone                │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">2048</span>)        │   <span style=\"color: #00af00; text-decoration-color: #00af00\">2,506,172,416</span> │ padding_mask[<span style=\"color: #00af00; text-decoration-color: #00af00\">0</span>][<span style=\"color: #00af00; text-decoration-color: #00af00\">0</span>],        │\n",
              "│ (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">GemmaBackbone</span>)               │                           │                 │ token_ids[<span style=\"color: #00af00; text-decoration-color: #00af00\">0</span>][<span style=\"color: #00af00; text-decoration-color: #00af00\">0</span>]            │\n",
              "├───────────────────────────────┼───────────────────────────┼─────────────────┼────────────────────────────┤\n",
              "│ token_embedding               │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">256000</span>)      │     <span style=\"color: #00af00; text-decoration-color: #00af00\">524,288,000</span> │ gemma_backbone[<span style=\"color: #00af00; text-decoration-color: #00af00\">0</span>][<span style=\"color: #00af00; text-decoration-color: #00af00\">0</span>]       │\n",
              "│ (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">ReversibleEmbedding</span>)         │                           │                 │                            │\n",
              "└───────────────────────────────┴───────────────────────────┴─────────────────┴────────────────────────────┘\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "\u001b[1m Total params: \u001b[0m\u001b[38;5;34m2,506,172,416\u001b[0m (9.34 GB)\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Total params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">2,506,172,416</span> (9.34 GB)\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "\u001b[1m Trainable params: \u001b[0m\u001b[38;5;34m2,506,172,416\u001b[0m (9.34 GB)\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">2,506,172,416</span> (9.34 GB)\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "\u001b[1m Non-trainable params: \u001b[0m\u001b[38;5;34m0\u001b[0m (0.00 B)\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Non-trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> (0.00 B)\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Gemma output:\n",
            "나랏말ᄊᆞ미ValueStyle.DoubleQuotes);\n",
            "            __tagHelperExecutionContext.AddHtmlAttribute(__tagHelperAttribute_2);\n",
            "            __tagHelperExecutionContext.AddHtmlAttribute(__tagHelperAttribute_3);\n",
            "            __tagHelperExecutionContext.AddHtmlAttribute(__tagHelperAttribute_4);\n",
            "            __tagHelperExecutionContext.AddHtmlAttribute(__tagHelperAttribute_5);\n",
            "            __tagHelperExecutionContext.AddHtmlAttribute(__tagHelperAttribute_6);\n",
            "            __tagHelperExecutionContext.AddHtmlAttribute(__tagHelperAttribute_7);\n",
            "            __tagHelperExecutionContext.AddHtmlAttribute(__tagHelperAttribute_8);\n",
            "            __tagHelperExecutionContext.AddHtmlAttribute(__tagHelperAttribute_9);\n",
            "            __tagHelperExecutionContext.AddHtmlAttribute(__tagHelperAttribute_10);\n",
            "            __tagHelperExecutionContext.AddHtmlAttribute(__tagHelperAttribute_11);\n",
            "            __tagHelperExecutionContext.AddHtmlAttribute(__tagHelperAttribute_12);\n",
            "            __tagHelperExecutionContext.AddHtmlAttribute(__tagHelperAttribute_13);\n",
            "            __tagHelperExecutionContext.AddHtmlAttribute(__tagHelperAttribute_14);\n",
            "            __tagHelperExecutionContext.AddHtmlAttribute(__tagHelperAttribute_15);\n",
            "            __tagHelperExecutionContext.AddHtmlAttribute(__tagHelperAttribute_16);\n",
            "            __tagHelperExecutionContext.AddHtmlAttribute(__tagHelperAttribute_17);\n",
            "            __tagHelperExecutionContext.AddHtmlAttribute(__tagHelperAttribute_18);\n",
            "            __tagHelperExecutionContext.AddHtmlAttribute(__tagHelperAttribute_19);\n",
            "\n",
            "TOTAL TIME ELAPSED: 22.37s\n",
            "\n",
            "Gemma output:\n",
            "나는월궁의션녀로셔츠\n",
            "\n",
            "1999년 1월 1일부터 2000년 12월 31일까지 11년간 1000여건의 헌법재판소 판례를 분석하여 헌법재판소의 헌법상의 판결과정을 살펴보았다. 헌법재판소는 헌법상의 판결과정을 헌법상의 판결과정으로 인식하고 있지 않다. 헌법재판소는 헌법상의 판결과정을 헌법상의 판결과정으로 인식하고 있지 않다. 헌법재판소는 헌법상의 판결과정을 헌법상의 판결과정으로 인식하고 있지 않다. 헌법재판소는 헌법상의 판결과정을 헌법상의 판결과정으로 인식하고 있지 않다. 헌법재판소는 헌법상의 판결과정을 헌법상의 판결과정으로 인식하고 있\n",
            "TOTAL TIME ELAPSED: 2.23s\n",
            "\n",
            "Gemma output:\n",
            "일쟝춘몽이라 불리는 <b>일장몽</b>(일장夢)은 1960년대에 일본에서 제작된 액션 영화이다.\n",
            "\n",
            "<h2>줄거리</h2>\n",
            "\n",
            "1960년대 일본에서 일장몽은 액션 영화로서 흥행을 하였다.\n",
            "\n",
            "<h2>출연</h2>\n",
            "\n",
            "* <b>오카다 쇼헤이</b>\n",
            "* <b>사카모토 켄지</b>\n",
            "* <b>사카모토 켄지</b>\n",
            "* <b>사카모토 켄지</b>\n",
            "* <b>사카모토 켄지</b>\n",
            "* <b>사카모토 켄지</b>\n",
            "* <b>사카모토 켄지</b>\n",
            "* <b>사카모토 켄지</b>\n",
            "* <b>사카모토 켄지</b>\n",
            "* <b>사카모토 켄지</b>\n",
            "* <b>사카모토 켄지</b>\n",
            "* <b>사카모토 켄지</b>\n",
            "* <b>사카모토 켄지</b>\n",
            "* <b>사카모토 켄지</b>\n",
            "* <b>사카모토 켄지\n",
            "TOTAL TIME ELAPSED: 2.24s\n",
            "\n",
            "Gemma output:\n",
            "션녀왈, 2009).\n",
            "\n",
            "  \\item 2009년 12월 1일부터 2010년 11월 30일까지 1년간 10개의 학교에서 100명의 학생을 대상으로 한 연구에서 학교폭력에 대한 학생들의 인식을 조사한 결과, 학교폭력에 대한 학생들의 인식은 학교폭력에 대한 인식이 높을수록 학교폭력에 대한 인식이 낮았다. 또한, 학교폭력에 대한 학생들의 인식은 학교폭력에 대한 인식이 높을수록 학교폭력에 대한 인식이 낮았다. 또한, 학교폭력에 대한 학생들의 인식은 학교폭력에 대한 인식이 높을수록 학교폭력에 대한 인식이 낮았다. 또한, 학교폭력에 대한 학생들의 인식은 학교폭력에 대한 인식이 높을수록 학교폭력에 대한 인식이 낮았다. 또한,\n",
            "TOTAL TIME ELAPSED: 2.27s\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "# LoRA Fine-tuning"
      ],
      "metadata": {
        "id": "G9kTUHx1AkrI"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "# Enable LoRA for the model and set the LoRA rank (4, 8 or 16).\n",
        "gemma.backbone.enable_lora(rank=lora_rank)\n",
        "gemma.summary()\n",
        "\n",
        "# Limit the input sequence length to 128 (to control memory usage).\n",
        "gemma.preprocessor.sequence_length = 1024\n",
        "# Use AdamW (a common optimizer for transformer models).\n",
        "optimizer = keras.optimizers.AdamW(\n",
        "    learning_rate=lr_value,\n",
        "    weight_decay=0.01,\n",
        ")\n",
        "# Exclude layernorm and bias terms from decay.\n",
        "optimizer.exclude_from_weight_decay(var_names=[\"bias\", \"scale\"])\n",
        "\n",
        "gemma.compile(\n",
        "    loss=keras.losses.SparseCategoricalCrossentropy(from_logits=True),\n",
        "    optimizer=optimizer,\n",
        "    weighted_metrics=[keras.metrics.SparseCategoricalAccuracy()],\n",
        ")\n"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 385
        },
        "id": "KTPbb1oFAmrT",
        "outputId": "3782e0ed-4a16-434b-a0d2-f376c6e8ed31"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "\u001b[1mPreprocessor: \"gemma_causal_lm_preprocessor\"\u001b[0m\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\">Preprocessor: \"gemma_causal_lm_preprocessor\"</span>\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓\n",
              "┃\u001b[1m \u001b[0m\u001b[1mTokenizer (type)                                  \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m                                            Vocab #\u001b[0m\u001b[1m \u001b[0m┃\n",
              "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┩\n",
              "│ gemma_tokenizer (\u001b[38;5;33mGemmaTokenizer\u001b[0m)                   │                                             \u001b[38;5;34m256,000\u001b[0m │\n",
              "└────────────────────────────────────────────────────┴─────────────────────────────────────────────────────┘\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓\n",
              "┃<span style=\"font-weight: bold\"> Tokenizer (type)                                   </span>┃<span style=\"font-weight: bold\">                                             Vocab # </span>┃\n",
              "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┩\n",
              "│ gemma_tokenizer (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">GemmaTokenizer</span>)                   │                                             <span style=\"color: #00af00; text-decoration-color: #00af00\">256,000</span> │\n",
              "└────────────────────────────────────────────────────┴─────────────────────────────────────────────────────┘\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "\u001b[1mModel: \"gemma_causal_lm\"\u001b[0m\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\">Model: \"gemma_causal_lm\"</span>\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓\n",
              "┃\u001b[1m \u001b[0m\u001b[1mLayer (type)                 \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1mOutput Shape             \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m        Param #\u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1mConnected to              \u001b[0m\u001b[1m \u001b[0m┃\n",
              "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━┩\n",
              "│ padding_mask (\u001b[38;5;33mInputLayer\u001b[0m)     │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;45mNone\u001b[0m)              │               \u001b[38;5;34m0\u001b[0m │ -                          │\n",
              "├───────────────────────────────┼───────────────────────────┼─────────────────┼────────────────────────────┤\n",
              "│ token_ids (\u001b[38;5;33mInputLayer\u001b[0m)        │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;45mNone\u001b[0m)              │               \u001b[38;5;34m0\u001b[0m │ -                          │\n",
              "├───────────────────────────────┼───────────────────────────┼─────────────────┼────────────────────────────┤\n",
              "│ gemma_backbone                │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m2048\u001b[0m)        │   \u001b[38;5;34m2,508,900,352\u001b[0m │ padding_mask[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m0\u001b[0m],        │\n",
              "│ (\u001b[38;5;33mGemmaBackbone\u001b[0m)               │                           │                 │ token_ids[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m0\u001b[0m]            │\n",
              "├───────────────────────────────┼───────────────────────────┼─────────────────┼────────────────────────────┤\n",
              "│ token_embedding               │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m256000\u001b[0m)      │     \u001b[38;5;34m524,288,000\u001b[0m │ gemma_backbone[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m0\u001b[0m]       │\n",
              "│ (\u001b[38;5;33mReversibleEmbedding\u001b[0m)         │                           │                 │                            │\n",
              "└───────────────────────────────┴───────────────────────────┴─────────────────┴────────────────────────────┘\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓\n",
              "┃<span style=\"font-weight: bold\"> Layer (type)                  </span>┃<span style=\"font-weight: bold\"> Output Shape              </span>┃<span style=\"font-weight: bold\">         Param # </span>┃<span style=\"font-weight: bold\"> Connected to               </span>┃\n",
              "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━┩\n",
              "│ padding_mask (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">InputLayer</span>)     │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>)              │               <span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> │ -                          │\n",
              "├───────────────────────────────┼───────────────────────────┼─────────────────┼────────────────────────────┤\n",
              "│ token_ids (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">InputLayer</span>)        │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>)              │               <span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> │ -                          │\n",
              "├───────────────────────────────┼───────────────────────────┼─────────────────┼────────────────────────────┤\n",
              "│ gemma_backbone                │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">2048</span>)        │   <span style=\"color: #00af00; text-decoration-color: #00af00\">2,508,900,352</span> │ padding_mask[<span style=\"color: #00af00; text-decoration-color: #00af00\">0</span>][<span style=\"color: #00af00; text-decoration-color: #00af00\">0</span>],        │\n",
              "│ (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">GemmaBackbone</span>)               │                           │                 │ token_ids[<span style=\"color: #00af00; text-decoration-color: #00af00\">0</span>][<span style=\"color: #00af00; text-decoration-color: #00af00\">0</span>]            │\n",
              "├───────────────────────────────┼───────────────────────────┼─────────────────┼────────────────────────────┤\n",
              "│ token_embedding               │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">256000</span>)      │     <span style=\"color: #00af00; text-decoration-color: #00af00\">524,288,000</span> │ gemma_backbone[<span style=\"color: #00af00; text-decoration-color: #00af00\">0</span>][<span style=\"color: #00af00; text-decoration-color: #00af00\">0</span>]       │\n",
              "│ (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">ReversibleEmbedding</span>)         │                           │                 │                            │\n",
              "└───────────────────────────────┴───────────────────────────┴─────────────────┴────────────────────────────┘\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "\u001b[1m Total params: \u001b[0m\u001b[38;5;34m2,508,900,352\u001b[0m (9.35 GB)\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Total params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">2,508,900,352</span> (9.35 GB)\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "\u001b[1m Trainable params: \u001b[0m\u001b[38;5;34m2,727,936\u001b[0m (10.41 MB)\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">2,727,936</span> (10.41 MB)\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "\u001b[1m Non-trainable params: \u001b[0m\u001b[38;5;34m2,506,172,416\u001b[0m (9.34 GB)\n"
            ],
            "text/html": [
              "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Non-trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">2,506,172,416</span> (9.34 GB)\n",
              "</pre>\n"
            ]
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "## multi epoch training"
      ],
      "metadata": {
        "id": "krYT30_aJhmj"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "history = gemma.fit(train, epochs=20, batch_size=2)\n",
        "gemma.backbone.save_lora_weights(f\"/content/drive/MyDrive/{lora_name}_{lora_rank}.lora.h5\")\n",
        "text_gen(\"나랏말ᄊᆞ미\")\n",
        "text_gen(\"나는월궁의션녀로셔\")\n",
        "text_gen(\"일쟝춘몽이라\")\n",
        "text_gen(\"션녀왈\")\n",
        "\n",
        "import matplotlib.pyplot as plt\n",
        "plt.plot(history.history['loss'])\n",
        "plt.show()"
      ],
      "metadata": {
        "id": "OdwDOfmxA03b",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "d0cc7d01-f6fb-49ca-db9a-b1b379995da6"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/20\n",
            "\u001b[1m122/122\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m79s\u001b[0m 236ms/step - loss: 0.7139 - sparse_categorical_accuracy: 0.3134\n",
            "Epoch 2/20\n",
            "\u001b[1m122/122\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m29s\u001b[0m 237ms/step - loss: 0.5748 - sparse_categorical_accuracy: 0.3836\n",
            "Epoch 3/20\n",
            "\u001b[1m122/122\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m29s\u001b[0m 237ms/step - loss: 0.5362 - sparse_categorical_accuracy: 0.4158\n",
            "Epoch 4/20\n",
            "\u001b[1m122/122\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m29s\u001b[0m 238ms/step - loss: 0.5061 - sparse_categorical_accuracy: 0.4435\n",
            "Epoch 5/20\n",
            "\u001b[1m122/122\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m29s\u001b[0m 238ms/step - loss: 0.4818 - sparse_categorical_accuracy: 0.4607\n",
            "Epoch 6/20\n",
            "\u001b[1m122/122\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m29s\u001b[0m 238ms/step - loss: 0.4673 - sparse_categorical_accuracy: 0.4719\n",
            "Epoch 7/20\n",
            "\u001b[1m122/122\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m29s\u001b[0m 237ms/step - loss: 0.4537 - sparse_categorical_accuracy: 0.4825\n",
            "Epoch 8/20\n",
            "\u001b[1m122/122\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m29s\u001b[0m 238ms/step - loss: 0.4361 - sparse_categorical_accuracy: 0.4963\n",
            "Epoch 9/20\n",
            "\u001b[1m122/122\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m29s\u001b[0m 238ms/step - loss: 0.4148 - sparse_categorical_accuracy: 0.5178\n",
            "Epoch 10/20\n",
            "\u001b[1m122/122\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m29s\u001b[0m 238ms/step - loss: 0.4018 - sparse_categorical_accuracy: 0.5296\n",
            "Epoch 11/20\n",
            "\u001b[1m122/122\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m29s\u001b[0m 238ms/step - loss: 0.3892 - sparse_categorical_accuracy: 0.5431\n",
            "Epoch 12/20\n",
            "\u001b[1m122/122\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m29s\u001b[0m 238ms/step - loss: 0.3736 - sparse_categorical_accuracy: 0.5566\n",
            "Epoch 13/20\n",
            "\u001b[1m122/122\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m29s\u001b[0m 238ms/step - loss: 0.3575 - sparse_categorical_accuracy: 0.5682\n",
            "Epoch 14/20\n",
            "\u001b[1m122/122\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m29s\u001b[0m 238ms/step - loss: 0.3424 - sparse_categorical_accuracy: 0.5818\n",
            "Epoch 15/20\n",
            "\u001b[1m122/122\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m29s\u001b[0m 238ms/step - loss: 0.3281 - sparse_categorical_accuracy: 0.5969\n",
            "Epoch 16/20\n",
            "\u001b[1m122/122\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m29s\u001b[0m 238ms/step - loss: 0.3167 - sparse_categorical_accuracy: 0.6052\n",
            "Epoch 17/20\n",
            "\u001b[1m122/122\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m29s\u001b[0m 238ms/step - loss: 0.3043 - sparse_categorical_accuracy: 0.6153\n",
            "Epoch 18/20\n",
            "\u001b[1m122/122\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m29s\u001b[0m 238ms/step - loss: 0.2961 - sparse_categorical_accuracy: 0.6267\n",
            "Epoch 19/20\n",
            "\u001b[1m122/122\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m29s\u001b[0m 238ms/step - loss: 0.2907 - sparse_categorical_accuracy: 0.6308\n",
            "Epoch 20/20\n",
            "\u001b[1m122/122\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m29s\u001b[0m 238ms/step - loss: 0.2760 - sparse_categorical_accuracy: 0.6457\n",
            "\n",
            "Gemma output:\n",
            "나랏말ᄊᆞ미. 길동이 젹상의 팔도의 슈명을 갑자기 믓�� shutil 팔도 막돌ᄒᆞ며, 졔장 ᄃᆡ감과 의논 갑위예 왈,\n",
            "TOTAL TIME ELAPSED: 15.11s\n",
            "\n",
            "Gemma output:\n",
            "나는 월궁의 션녀로셔. 월국이 ᄯᅥ나난 후의, 슈쳔업거셔 궁실 소셔 젼ᄒᆞ여 쳔인의 졀을 졀의져 ᄒᆞ거날, 길동이 쳔인의 졀을 쳔지지 못ᄒᆞ고 졔인의 졀을 졀의지지 못ᄒᆞ며, 션업난 후의 슈업거ᄅᆞᆯ 쳔인의 졔을 지지 못ᄒᆞ고, 션스러ᄂᆞᆫ 션업난 후의 션업거ᄂᆞᆫ 션스러워 ᄒᆞ고 슈염난 션스러워 슈을 쳔지ᄒᆞ고, 궁실 소셔 션업난 후의 션업거ᄂᆞᆫ 션스러워 궁장의 쳔지지 못ᄒᆞ고, 궁장난 션장미 졔을 지지 못ᄒᆞ고 션장졀\n",
            "TOTAL TIME ELAPSED: 2.49s\n",
            "\n",
            "Gemma output:\n",
            "일쟝춘몽이라 ᄂᆡ의 몽 chu chu ᄭ�<h1>지지원 (순우군 셔리 지구}`;\n",
            "    \n",
            "    let ngwgi = new Date().getTime();\n",
            "    let ngwgi2 = new Date(ngwgi - 10000);\n",
            "    let ngwgi3 = new Date(ngwgi - 15000);\n",
            "    \n",
            "    let ngwgi4 = new Date(ngwgi - 10000);\n",
            "    let ngwgi5 = new Date(ngwgi - 15000);\n",
            "    \n",
            "    let ngwgi6 = new Date(ngwgi - 10000);\n",
            "    let ngwgi7 = new Date(ngwgi - 15000);\n",
            "    \n",
            "    let ngwgi8 = new Date(ngwgi - 10000);\n",
            "    let ngwgi9 = new Date(ngwgi - 15000);\n",
            "    \n",
            "    let ngwgi10 = new Date(ngwgi - 1000\n",
            "TOTAL TIME ELAPSED: 2.51s\n",
            "\n",
            "Gemma output:\n",
            "션녀왈 ᄒᆞᆫ 졔이 졈쳔의 졔녀로 ᄒᆞ와 ᄇᆡᆨ씨 왈 ᄲᅮ 졔녀라. 길동이 션화을 졀화 ᄒᆞ시고 졀젹이 슈아 졀젹 왈,\n",
            "TOTAL TIME ELAPSED: 0.89s\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 640x480 with 1 Axes>"
            ],
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAiwAAAGdCAYAAAAxCSikAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjcuMSwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/bCgiHAAAACXBIWXMAAA9hAAAPYQGoP6dpAABKSUlEQVR4nO3de1xUdf4/8NfMwMzIHUWGiyM3FdQUFJTwbpJgltpNbSuVTHfN2loqL9/dtLLfUta2bmXqul5z17SbueliOYlXBAVN84KCXJXhKne5zZzfH+gUCcogcObyej4e5/GIM5/z8X06jbw65/P5HIkgCAKIiIiITJhU7AKIiIiI7oaBhYiIiEweAwsRERGZPAYWIiIiMnkMLERERGTyGFiIiIjI5DGwEBERkcljYCEiIiKTZyN2AR1Br9fj2rVrcHR0hEQiEbscIiIiagNBEFBZWQkvLy9IpXe+h2IRgeXatWtQq9Vil0FERETtkJubi169et2xjUUEFkdHRwBNJ+zk5CRyNURERNQWFRUVUKvVht/jd2IRgeXWYyAnJycGFiIiIjPTluEcHHRLREREJo+BhYiIiEweAwsRERGZPAYWIiIiMnkMLERERGTyGFiIiIjI5DGwEBERkcljYCEiIiKTx8BCREREJo+BhYiIiEweAwsRERGZPAYWIiIiMnkMLHdQVlOPtQczsOjLn8QuhYiIyKoxsNyBIADvxV/EzpN5yC+/IXY5REREVouB5Q5c7eUIUbsAABLSisQthoiIyIoxsNzF+EB3AMCBi4UiV0JERGS9GFju4lZgOZpejLpGncjVEBERWScGlrsY6OUENwcFqut1OJl1XexyiIiIrBIDy11IpRKMC+wJAEhI42MhIiIiMTCwtMGtwHKAA2+JiIhEwcDSBqP79IRMKkF6YRVyS2vELoeIiMjqMLC0gbOdLUJ7uwLgYyEiIiIxMLC00bigW+NY+FiIiIioqzGwtNG4fjenN2cUo7aB05uJiIi6EgNLG/X3dITKSYHaBj2SMkvFLoeIiMiqMLC0kUQi4aq3REREImFgMcK4m4Hl4CWOYyEiIupKDCxGGNmnB2xlEmQWVyOzuFrscoiIiKwGA4sRHJW2CPPpDoDTm4mIiLoSA4uRxgdx1VsiIqKuxsBipFsDb49fKcGNek5vJiIi6goMLEbq4+4Ab5duqG/UI/FKsdjlEBERWYV2BZbVq1fD19cXSqUS4eHhSE5OvmP7srIyLFy4EJ6enlAoFOjXrx/27t17T32KRSKR/PJY6CIfCxEREXUFowPLjh07EBsbi+XLlyM1NRXBwcGIiopCYWHLg1Dr6+vx4IMPIisrC19++SXS0tKwfv16eHt7t7tPsd1a9fZAWiEEQRC5GiIiIssnEYz8jRseHo5hw4bhk08+AQDo9Xqo1Wq89NJLWLJkyW3t165di/fffx8XL16Era1th/T5WxUVFXB2dkZ5eTmcnJyMOZ12qalvRMhbP6Bep8f+2DHo4+7Y6X8mERGRpTHm97dRd1jq6+uRkpKCyMjIXzqQShEZGYnExMQWj9m9ezciIiKwcOFCqFQq3HffffjrX/8KnU7X7j7r6upQUVHRbOtKdnIbhPvfmt7Mx0JERESdzajAUlxcDJ1OB5VK1Wy/SqWCVqtt8ZgrV67gyy+/hE6nw969e/HGG2/gb3/7G95555129xkXFwdnZ2fDplarjTmNDmFYpp/rsRAREXW6Tp8lpNfr4e7ujn/+858IDQ3FjBkz8Oc//xlr165td59Lly5FeXm5YcvNze3AittmXGDTwNvkzFJU1TV2+Z9PRERkTWyMaezm5gaZTIaCgoJm+wsKCuDh4dHiMZ6enrC1tYVMJjPs69+/P7RaLerr69vVp0KhgEKhMKb0DufnZg+fHnbILqnB0fRiRA1suVYiIiK6d0bdYZHL5QgNDYVGozHs0+v10Gg0iIiIaPGYkSNHIj09HXq93rDv0qVL8PT0hFwub1efpuDXb2/mOBYiIqLOZfQjodjYWKxfvx5btmzBhQsXsGDBAlRXVyMmJgYAMGvWLCxdutTQfsGCBSgtLcXLL7+MS5cuYc+ePfjrX/+KhQsXtrlPU3XrsVACpzcTERF1KqMeCQHAjBkzUFRUhGXLlkGr1SIkJATx8fGGQbM5OTmQSn/JQWq1Gvv27cOf/vQnDB48GN7e3nj55ZexePHiNvdpqu737wGlrRT55bVIK6hEkEfnT6kmIiKyRkavw2KKunodll+L2ZSMA2lFWBwdhAXjArr0zyYiIjJnnbYOC91ufBCnNxMREXU2BpZ7dGuZ/pTs66iobRC5GiIiIsvEwHKPevewQ0BPe+j0Ao5c5tubiYiIOgMDSwcwrHp7kY+FiIiIOgMDSwcYd2s9lktF0OvNfgwzERGRyWFg6QDD/FxhJ5ehqLIO5/O79kWMRERE1oCBpQMobGQY2ccNQNMickRERNSxGFg6yC9vb+Yy/URERB2NgaWD3Fqm/1TOdVyvrhe5GiIiIsvCwNJBvFy6IVDlCL0AHLrMuyxEREQdiYGlA40LarrLcpCPhYiIiDoUA0sHGs/pzURERJ2CgaUDhfq4wlFhg9Lqepy5Wi52OURERBaDgaUD2cqkGNW3aXozV70lIiLqOAwsHezXj4WIiIioYzCwdLCxN6c3n8krQ3FVncjVEBERWQYGlg6mclJioJcTBAE4xLssREREHYKBpRNw1VsiIqKOxcDSCW6tenvoUhF0nN5MRER0zxhYOkGI2gXO3WxRfqMBp3Ovi10OERGR2WNg6QQ2MinG9Gu6y3LgIh8LERER3SsGlk4y/uZjoQNpXI+FiIjoXjGwdJIx/XpCIgHOXatAYUWt2OUQERGZNQaWTuLmoMBgb2cAXESOiIjoXjGwdKJxt1a95WMhIiKie8LA0onGBzUFlsOXitGg04tcDRERkfliYOlEg72d0cNejsq6RqRkc3ozERFRezGwdCKpVGKY3pzAVW+JiIjajYGlk91a9ZbjWIiIiNqPgaWTjenbE1IJcFFbiWtlN8Quh4iIyCwxsHQyV3s5hvR2BcDHQkRERO3FwNIFxvOxEBER0T1hYOkCt9ZjOZpejLpGncjVEBERmR8Gli4wwNMJPR0VqK7X4WQWpzcTEREZi4GlC0ilEowzvL2Zj4WIiIiM1a7Asnr1avj6+kKpVCI8PBzJycmttt28eTMkEkmzTalUNmszZ86c29pER0e3pzSTdWvVW769mYiIyHg2xh6wY8cOxMbGYu3atQgPD8eqVasQFRWFtLQ0uLu7t3iMk5MT0tLSDD9LJJLb2kRHR2PTpk2GnxUKhbGlmbRRfd0gk0qQUVSN3NIaqLvbiV0SERGR2TD6DsuHH36IefPmISYmBgMGDMDatWthZ2eHjRs3tnqMRCKBh4eHYVOpVLe1USgUzdq4uroaW5pJc1LaItTn1vRm3mUhIiIyhlGBpb6+HikpKYiMjPylA6kUkZGRSExMbPW4qqoq+Pj4QK1WY+rUqTh37txtbRISEuDu7o7AwEAsWLAAJSUlrfZXV1eHioqKZps5GB9467EQ12MhIiIyhlGBpbi4GDqd7rY7JCqVClqttsVjAgMDsXHjRnz77bfYtm0b9Ho9RowYgby8PEOb6OhobN26FRqNBu+99x4OHjyISZMmQadreQpwXFwcnJ2dDZtarTbmNEQzPqhp4O2xjGLUNnB6MxERUVsZPYbFWBEREYiIiDD8PGLECPTv3x/r1q3DihUrAAAzZ840fD5o0CAMHjwYAQEBSEhIwIQJE27rc+nSpYiNjTX8XFFRYRahJVDlCE9nJfLLa5GUWYqxN2cOERER0Z0ZdYfFzc0NMpkMBQUFzfYXFBTAw8OjTX3Y2tpiyJAhSE9Pb7WNv78/3NzcWm2jUCjg5OTUbDMHEonE8DJETm8mIiJqO6MCi1wuR2hoKDQajWGfXq+HRqNpdhflTnQ6Hc6ePQtPT89W2+Tl5aGkpOSObczVrVVvOfCWiIio7YyeJRQbG4v169djy5YtuHDhAhYsWIDq6mrExMQAAGbNmoWlS5ca2r/99tv4/vvvceXKFaSmpuKZZ55BdnY2nn/+eQBNA3Jff/11HD9+HFlZWdBoNJg6dSr69OmDqKioDjpN0zGyjxtsZRJkldQgs7ha7HKIiIjMgtFjWGbMmIGioiIsW7YMWq0WISEhiI+PNwzEzcnJgVT6Sw66fv065s2bB61WC1dXV4SGhuLYsWMYMGAAAEAmk+HMmTPYsmULysrK4OXlhYkTJ2LFihUWtxYLADgobDDcrzuOppfgwMVC+I3yE7skIiIikycRBEEQu4h7VVFRAWdnZ5SXl5vFeJZ/Hb6Cd/ZcwJh+PbH1ueFil0NERCQKY35/811CIrg18Pb4lRLU1DeKXA0REZHpY2ARQUBPB/Ry7Yb6Rj0SM1pfII+IiIiaMLCIQCKR/GrVW84WIiIiuhsGFpHcWvU2Ia0IFjCMiIiIqFMxsIgkwt8Nchsp8q7fQEZRldjlEBERmTQGFpF0k8twv38PAMCBi3wZIhER0Z0wsIho/M3ZQl+fuooGnV7kaoiIiEwXA4uIJg/2hJPSBhfyK/D3Hy6JXQ4REZHJYmARkbujEu8+PhgAsOZgBo6lF4tcERERkWliYBHZQ4M8MXOYGoIA/GnnaZRW14tdEhERkclhYDEByx4ZgICe9iioqMOiL3/iNGciIqLfYGAxAXZyG3z81FDIZVLsv1CIz45ni10SERGRSWFgMREDvJywZFIQAOCdPRdwUVshckVERESmg4HFhMSM9MUDQe6ob9Tjpf+cwo16ndglERERmQQGFhMikUjw/hOD0dNRgcuFVXhnz3mxSyIiIjIJDCwmpoeDAh9ODwYA/DspB/E/a0WuiIiISHwMLCZodN+e+P0YfwDA4q/O4FrZDZErIiIiEhcDi4l6dWIgBvdyRvmNBvxpx2no9JzqTERE1ouBxUTJbaT4aOYQ2MtlSMosxacH0sUuiYiISDQMLCbM180eb0+9DwCwSnMZKdmlIldEREQkDgYWE/fYUG9MC/GCTi/gj9tPo/xGg9glERERdTkGFhMnkUiwYtp96N3dDlfLbuDP35zl0v1ERGR1GFjMgKPSFv+YGQIbqQTfncnHFyfzxC6JiIioSzGwmIkhvV3xpwf7AQCW7z6HjKIqkSsiIiLqOgwsZuQPYwMwIqAHbjTo8Mftp1DXyKX7iYjIOjCwmBGZVIK/zwiBq50tzl2rwMr4NLFLIiIi6hIMLGZG5aTE+080Ld2/4UgmEtIKRa6IiIio8zGwmKHIASrMjvABALz2xU8orKwVuSIiIqLOxcBippY+1B9BHo4orqrHqzt/gp5L9xMRkQVjYDFTSlsZPn5qCBQ2Uhy+XIwNRzLFLomIiKjTMLCYsb4qRyx7ZAAAYOW+izibVy5yRURERJ2DgcXM/W54b0QP9ECDTsAfPz+F6rpGsUsiIiLqcAwsZk4ikeDdxwfB01mJzOJqLN99TuySiIiIOhwDiwVwsZPj7zNCIJEAX6bk4dvTV8UuiYiIqEMxsFiI+/174KXxfQAAf/nmZ+SW1ohcERERUcdpV2BZvXo1fH19oVQqER4ejuTk5Fbbbt68GRKJpNmmVCqbtREEAcuWLYOnpye6deuGyMhIXL58uT2lWbU/TuiLUB9XVNY14o+fn0KDTi92SURERB3C6MCyY8cOxMbGYvny5UhNTUVwcDCioqJQWNj6iqtOTk7Iz883bNnZ2c0+X7lyJT766COsXbsWSUlJsLe3R1RUFGpruSCaMWxkUqyaEQJHpQ1O5ZThH/sZ+oiIyDIYHVg+/PBDzJs3DzExMRgwYADWrl0LOzs7bNy4sdVjJBIJPDw8DJtKpTJ8JggCVq1ahb/85S+YOnUqBg8ejK1bt+LatWvYtWtXu07Kmqm72+Gvjw4CAKxOSMexjGKRKyIiIrp3RgWW+vp6pKSkIDIy8pcOpFJERkYiMTGx1eOqqqrg4+MDtVqNqVOn4ty5X2ayZGZmQqvVNuvT2dkZ4eHhrfZZV1eHioqKZhv94pFgL0wP6wVBAP604zSuV9eLXRIREdE9MSqwFBcXQ6fTNbtDAgAqlQparbbFYwIDA7Fx40Z8++232LZtG/R6PUaMGIG8vDwAMBxnTJ9xcXFwdnY2bGq12pjTsApvThkIfzd7FFTU4fefpaC2QSd2SURERO3W6bOEIiIiMGvWLISEhGDs2LH4+uuv0bNnT6xbt67dfS5duhTl5eWGLTc3twMrtgx2cht8+sxQOCpskJxVihf/cwqNHIRLRERmyqjA4ubmBplMhoKCgmb7CwoK4OHh0aY+bG1tMWTIEKSnpwOA4Thj+lQoFHBycmq20e2CPJzwr9lhkNtIsf9CAf7vm7MQBL4kkYiIzI9RgUUulyM0NBQajcawT6/XQ6PRICIiok196HQ6nD17Fp6engAAPz8/eHh4NOuzoqICSUlJbe6TWhfu3wMfPzUEUgmw82Qe3t+XJnZJRERERjP6kVBsbCzWr1+PLVu24MKFC1iwYAGqq6sRExMDAJg1axaWLl1qaP/222/j+++/x5UrV5CamopnnnkG2dnZeP755wE0zSB65ZVX8M4772D37t04e/YsZs2aBS8vL0ybNq1jztLKRQ30MMwc+jQhAxv5ZmciIjIzNsYeMGPGDBQVFWHZsmXQarUICQlBfHy8YdBsTk4OpNJfctD169cxb948aLVauLq6IjQ0FMeOHcOAAQMMbRYtWoTq6mrMnz8fZWVlGDVqFOLj429bYI7ab+bw3iiuqsMH31/C29+dRw8HOaaGeItdFhERUZtIBAsY1FBRUQFnZ2eUl5dzPMsdCIKAt/57HpuPZcFGKsGGOcMwtl9PscsiIiIrZczvb75LyIpIJBIse3gAHgn2QqNewIJtKTidWyZ2WURERHfFwGJlpFIJ/vZkMEb3dUNNvQ4xm5KRXlgldllERER3xMBiheQ2Uqx5JhTBvZxxvaYBszcmI7/8hthlERERtYqBxUo5KGywcc4w+LvZ42rZDczemIyyGi7hT0REpomBxYr1cFBg69zhUDkpcKmgCnO3nMSNei7hT0REpoeBxcr1crXD1ufC4aS0QUr2dSz8TyoauIQ/ERGZGAYWQqCHIzbMGQaFjRQ/XizEkq+4hD8REZkWBhYCAAzz7Y7VvxsKmVSCr1Lz8O7/LopdEhERkQEDCxlEDlAh7rGmJfzXHbqC9YeuiFwRERFREwYWamZ6mBqLo4MAAP9v7wV8lZInckVEREQMLNSCP4z1x9xRfgCARV+dwYGLhSJXRERE1o6BhW4jkUjw54f649Eh3tDpBSz4dwpSsq+LXRYREVkxBhZqkVQqwconBmNsv56obdDjuc0ncLmgUuyyiIjISjGwUKtsZVKseWYoQtQuKL/RgFkbk3GtjEv4ExFR12NgoTuyk9tg05xh6OPugPzyWjy7IQnXq7mEPxERdS0GFrorV3s5tj43HJ7OSmQUVSNm8wnU1DeKXRYREVkRBhZqEy+Xbtj63HC42NnidG4ZFmzjEv5ERNR1GFiozfqqHLFh9jAobaU4eKkIr3/xE/R6LuFPRESdj4GFjBLq44o1T4dCJpVg1+lr+H97L/C9Q0RE1OkYWMho44Pc8f4TgwEAG45kYslXZ1HfyMdDRETUeRhYqF0eG9oLK6YOhFQC7DiZi2c2JKGUs4eIiKiTMLBQuz0b4YsNs4fBQWGD5MxSTFt9lIvLERFRp2BgoXsyPsgdX78wAuru3ZBTWoPHPj2GhDS+e4iIiDoWAwvds34qR3y7cBSG+3ZHZV0jntt8AhuPZHIwLhERdRgGFuoQ3e3l2PZ8OJ4M7QW9ALz93Xn83zc/c60WIiLqEAws1GHkNlKsfGIw/vxQf0gkwPbkHC7lT0REHYKBhTqURCLBvDH++NesMNjLZTh+pRTTPj2K9MIqsUsjIiIzxsBCnWJCfxW+emEEvF26IbukBo9+ehSHLhWJXRYREZkpBhbqNEEeTvj2xZEI83FFZW0jYjafwJZjWRyMS0RERmNgoU7l5qDAv+eF47Gh3tDpBSzffQ5vfMvBuEREZBwGFup0ChsZ/vZkMJZMCoJEAmw7noM5m5JRVsPBuERE1DYMLNQlJBIJ/jA2AOueCYWdXIaj6SV49NNjuFLEwbhERHR3DCzUpSYO9MCXf2gajJtZXI1pq4/iyOViscsiIiITx8BCXW6AlxN2LRyJob1dUFHbiNmbkvHZ8WyxyyIiIhPWrsCyevVq+Pr6QqlUIjw8HMnJyW067vPPP4dEIsG0adOa7Z8zZw4kEkmzLTo6uj2lkZno6ajAf+bdj0eHNA3GfWPXz1j27c9o5GBcIiJqgdGBZceOHYiNjcXy5cuRmpqK4OBgREVFobDwzi+8y8rKwmuvvYbRo0e3+Hl0dDTy8/MN2/bt240tjcyM0laGD6cH4/WoQADA1sRsxGw+gfIbDSJXRkREpsbowPLhhx9i3rx5iImJwYABA7B27VrY2dlh48aNrR6j0+nw9NNP46233oK/v3+LbRQKBTw8PAybq6ursaWRGZJIJFg4vg/WPhOKbrYyHL5cjEc/PYrM4mqxSyMiIhNiVGCpr69HSkoKIiMjf+lAKkVkZCQSExNbPe7tt9+Gu7s75s6d22qbhIQEuLu7IzAwEAsWLEBJSYkxpZGZi77PA1/8IQKezkpcKWoajHssg4NxiYioiVGBpbi4GDqdDiqVqtl+lUoFrVbb4jFHjhzBhg0bsH79+lb7jY6OxtatW6HRaPDee+/h4MGDmDRpEnQ6XYvt6+rqUFFR0Wwj83eftzO+XTgSIWoXlN9owKwNydh8NBM6PVfGJSKydp06S6iyshLPPvss1q9fDzc3t1bbzZw5E1OmTMGgQYMwbdo0fPfddzhx4gQSEhJabB8XFwdnZ2fDplarO+kMqKu5Oynx+fz7MSXYC416AW/+9zymrT6KlOxSsUsjIiIRGRVY3NzcIJPJUFBQ0Gx/QUEBPDw8bmufkZGBrKwsPPLII7CxsYGNjQ22bt2K3bt3w8bGBhkZGS3+Of7+/nBzc0N6enqLny9duhTl5eWGLTc315jTIBOntJXhHzND8OYjA+CosMHZq+V4fE0i/rTjNAoqasUuj4iIRGBUYJHL5QgNDYVGozHs0+v10Gg0iIiIuK19UFAQzp49i9OnTxu2KVOmYPz48Th9+nSrd0by8vJQUlICT0/PFj9XKBRwcnJqtpFlkUgkmDPSDwdeH4cZYWpIJMA3p65i/AcJWJOQgbrGlh8XEhGRZZIIRr46d8eOHZg9ezbWrVuH4cOHY9WqVdi5cycuXrwIlUqFWbNmwdvbG3FxcS0eP2fOHJSVlWHXrl0AgKqqKrz11lt4/PHH4eHhgYyMDCxatAiVlZU4e/YsFArFXWuqqKiAs7MzysvLGV4s1Jm8Mry5+xxSc8oAAL497PDGwwPwQJA7JBKJuMUREVG7GPP728bYzmfMmIGioiIsW7YMWq0WISEhiI+PNwzEzcnJgVTa9hs3MpkMZ86cwZYtW1BWVgYvLy9MnDgRK1asaFNYIeswuJcLvvzDCOw6fRXv/u8iskpqMHfLSYwL7Ik3Hh6AgJ4OYpdIRESdyOg7LKaId1isS1VdIz75MR0bjlxBg06AjVSC50b54aUH+sBRaSt2eURE1EbG/P5mYCGzlVlcjXe+Ow/NxaZVlt0cFFgcHYjHh/aCVMrHREREpo6BhazKgYuFePu784bVcYPVLnhrykCEqF3ELYyIiO6IgYWsTn2jHpuOZuIjzWVU1zfNIHoytBdejw6Eu6NS5OqIiKglDCxktQoravFefBq+Ss0DADgobPDyhL6YPcIXcptOXSeRiIiMxMBCVu9UznW8ufscfsorBwD497THsocHYFygu8iVERHRLQwsRAD0egFfpuZhZfxFFFfVAwAi+7vjL5MHwNfNXuTqiIiIgYXoVypqG/Cx5jI2Hc1Co16AXCbF3NF+eHF8H9grjF6KiIiIOggDC1EL0gur8PZ353HoUhEAwN1RgaUPBWFaiDdXyyUiEgEDC1ErBEGA5kIhVuw5j+ySGgBAqI8r3nxkIAb1cha5OiIi68LAQnQXdY06/OtwJlYfSEdNvQ4SCTBzmBqvTQxEDwe+EoKIqCswsBC1kba8Fu/+7wJ2nb4GAHBU2uBPkf3wbIQPbGWcBk1E1JkYWIiMdDKrFMt3n8O5axUAgL7uDnhzykCM7OMmcmVERJaLgYWoHXR6ATtP5uL9fWkorW6aBh090AN/ntwf6u52IldHRGR5GFiI7kF5TQP+vv8SPjueDZ1egMJGit+PDcCCsQHoJpeJXR4RkcVgYCHqAGnaSry5+xwSr5QAALyclfjz5AF4aJAHp0ETEXUABhaiDiIIAuJ/1uKdPRdwtewGAOB+/+54c8pABHnwvzUionvBwELUwW7U67DuUAbWJGSgrlEPqQR45n4fxD7YDy52crHLIyIySwwsRJ0k73oN/rr3Avae1QIAXO1s8erEQDw1vDdkUj4mIiIyBgMLUSc7ll6MN/97DpcKqgAA/T2d8NaUgRju113kyoiIzAcDC1EXaNTp8e+kHPzt+zRU1DYCAKYEe2HpQ0HwdO4mcnVERKaPgYWoC5VU1eFvP1zC9uQcCALQzVaGFx/og7mj/KC05TRoIqLWMLAQieDnq+V4c/c5nMy+DgDo3d0O//dQEKIGcho0EVFLGFiIRCIIAnb/dA1/3XsBBRV1AIDhvt3xl4f7Y3AvF3GLIyIyMQwsRCKrrmvEuoMZ+OfhK6ht0AMAHh3ijdejAuHlwvEtREQAA4vY5RAZXCu7gQ/2peHrU1cBAAobKeaP8ccfxgbAXmEjcnVEROJiYCEyMWfyyvDOdxeQnFUKAOjpqMBrE/vhiVA1128hIqvFwEJkggRBwL5zWsT97yKyS2oAAEEejvjL5AEY1ddN5OqIiLoeAwuRCatr1OGzxGx8pLlsWL9lQpA7lj7UH33cHUSujoio6zCwEJmB69X1+IfmMrYdz0ajXoBMKsHT4b3xSmQ/dLfn+4mIyPIxsBCZkYyiKsTtvYj9FwoAAI5KG7z0QB/MHuELhQ0XniMiy8XAQmSGjmUU453vLuB8fgUAQN29G5ZO6o9J93HhOSKyTAwsRGZKpxfwdWoe3t+XhsLKpoXnwnxc8ZeHByBE7SJucUREHYyBhcjMVdc14p+HrmDdoQzDwnNTQ7ywKDoI3lx4jogsBAMLkYXQltfi/X1p+Co1D0DTwnPPj/bDgnF94MCF54jIzDGwEFmYs3nleGfPeSRlNi085+Ygx+tRgZgepub4FiIyW8b8/pa25w9YvXo1fH19oVQqER4ejuTk5DYd9/nnn0MikWDatGnN9guCgGXLlsHT0xPdunVDZGQkLl++3J7SiCzSoF7O+Hz+/fjns6Hw7WGH4qp6LP7qLOZ/loKymnqxyyMi6nRGB5YdO3YgNjYWy5cvR2pqKoKDgxEVFYXCwsI7HpeVlYXXXnsNo0ePvu2zlStX4qOPPsLatWuRlJQEe3t7REVFoba21tjyiCyWRCLBxIEe+P5PY/F/DwVBLpPih/MFeOgfh5GSXSp2eUREncroR0Lh4eEYNmwYPvnkEwCAXq+HWq3GSy+9hCVLlrR4jE6nw5gxY/Dcc8/h8OHDKCsrw65duwA03V3x8vLCq6++itdeew0AUF5eDpVKhc2bN2PmzJl3rYmPhMga/Xy1HC/+JxVZJTWQSSWIfbAfFowNgJTvJiIiM9Fpj4Tq6+uRkpKCyMjIXzqQShEZGYnExMRWj3v77bfh7u6OuXPn3vZZZmYmtFptsz6dnZ0RHh7eap91dXWoqKhothFZm/u8nfHdH0djWogXdHoB7+9Lw+xNySi6OR2aiMiSGBVYiouLodPpoFKpmu1XqVTQarUtHnPkyBFs2LAB69evb/HzW8cZ02dcXBycnZ0Nm1qtNuY0iCyGg8IGf58RgpVPDIbSVorDl4sx6R+HceRysdilERF1qHYNum2ryspKPPvss1i/fj3c3DrubbRLly5FeXm5YcvNze2wvonMjUQiwfQwNf774igEqhxRXFWHZzcm4YN9aWjU6cUuj4ioQxi1kIObmxtkMhkKCgqa7S8oKICHh8dt7TMyMpCVlYVHHnnEsE+vb/oL1MbGBmlpaYbjCgoK4Onp2azPkJCQFutQKBRQKBTGlE5k8fqqHLFr4Ui8/d15bE/OwScH0pGUWYKPnhoCT2cuNkdE5s2oOyxyuRyhoaHQaDSGfXq9HhqNBhEREbe1DwoKwtmzZ3H69GnDNmXKFIwfPx6nT5+GWq2Gn58fPDw8mvVZUVGBpKSkFvskotZ1k8sQ99ggfPzUEDgobHAi6zom/eMw9p8vuPvBREQmzOilMmNjYzF79myEhYVh+PDhWLVqFaqrqxETEwMAmDVrFry9vREXFwelUon77ruv2fEuLi4A0Gz/K6+8gnfeeQd9+/aFn58f3njjDXh5ed22XgsRtc0jwV4Y3MsZL/7nFM5eLcfzW0/iuZF+WDIpCHKbTn0STETUKYwOLDNmzEBRURGWLVsGrVaLkJAQxMfHGwbN5uTkQCo17i/ERYsWobq6GvPnz0dZWRlGjRqF+Ph4KJVKY8sjopt8etjjywUReO9/adh4NBMbj2biZHYpPn5qCHx62ItdHhGRUbg0P5EV2H++AK99+RPKahrgoLBB3GOD8Eiwl9hlEZGV6/Sl+YnIvEQOUGHvH0cjzMcVVXWNeGn7KSz9+gxu1OvELo2IqE0YWIishJdLN3w+/368OL4PJBJge3Iupq4+gssFlWKXRkR0VwwsRFbERibFa1GB+Oy5cLg5KHCpoAqPfHIEO0/kwgKeDhORBWNgIbJCo/q64X8vj8bovm6obdBj0Vdn8MqO06iqaxS7NCKiFjGwEFmpno4KbIkZjtejAiGTSvDt6Wt4+KPD+PlqudilERHdhoGFyIpJpRIsHN8HO+bfDy9nJbJKavDYp8ew+WgmHxERkUlhYCEihPl2x96XRyOyvwr1Oj3e/O95zPjncXx7+ipqGziTiIjEx3VYiMhAEARsOpqFuP9dQIOu6a8GJ6UNpoZ4Y3qYGvd5O0EikYhcJRFZCmN+fzOwENFt8q7X4IuTefgyJQ9Xy24Y9gd5OGJ6mBrThniju71cxAqJyBIwsBBRh9DrBRzLKMHOk7mIP6dFfWPT29ZtZRJE9ldhepgao/u6wUbGp8tEZDwGFiLqcOU1Ddh95hq+OJmLM3m/zCRSOSnw+NBeeDJMDT83vqOIiNqOgYWIOtWF/Ap8cTIP35zKw/WaBsP+Yb6ueDJMjcmDPGGvMPrdqkRkZRhYiKhL1DfqoblQgJ0nc3HwUhH0N/82sZPL8PBgT0wPUyPUx5UDdYmoRQwsRNTltOW1+PpUHr44mYfM4mrDfn83ezwZpsZjQ72hclKKWCERmRoGFiISjSAIOJl9HTtP5GLP2XzU3HwjtFQCjAt0x/SwXnggSAW5DQfqElk7BhYiMgnVdY3YczYfX5zMxYms64b93e3liBnhi7mj/WAn51gXImvFwEJEJiejqApfpuThq5Q8FFbWAQDcHBR4ObIvZg5Tw5ZTo4msDgMLEZmsRp0ee87m42/fX0JOaQ0AwLeHHV6LCsRD93lCKuUAXSJrwcBCRCavvlGPz0/k4CPNZRRX1QMABvdyxuLoIIzs4yZydUTUFRhYiMhsVNU1YsPhTPzzUAaqbw7QHd3XDYujg3Cft7PI1RFRZ2JgISKzU1xVh09+TMe/k7INL16cEuyFVyf2g08PrqBLZIkYWIjIbOWU1OBvP6Th29PXAAA2UgmeDu+Nlyb0hZuDQuTqiKgjMbAQkdn7+Wo5Vu5Lw6FLRQAAe7kMz4/2x7wx/nDgsv9EFoGBhYgsxrH0Yrwbf9HwwsUe9nK89EAf/C7ch4vPEZk5BhYisiiCIGDvWS0++D7NsOx/7+52eHViPzwy2ItToYnMFAMLEVmkBp0eO0/mYtX+yyi6ufjcAE8nLJ4UhDF93fiSRSIzw8BCRBatpr4RG49kYu3BK6iqawQAjAjogcXRQQhWu4hbHBG1GQMLEVmF0up6rD6Qjs8Ss1Gv0wMAJg/yxGtRgfBz41RoIlPHwEJEViXveg0+/OESvjl1FYIAyKQSTA9T448T+sDTuZvY5RFRKxhYiMgqXcivwPv70vDjxUIAgNxGitkRPlgwrg+628tFro6IfouBhYisWnJmKd7fdxEnsq4DABwUNnh+tB+eH801XIhMCQMLEVk9QRCQcKkI78en4Xx+BQCgu70cL4wLwDP3+0BpKxO5QiJiYCEiukmvF7D353x8+P0lXLm5hounsxIvT+iLJ0J7wUbGxeeIxMLAQkT0G406Pb5KzcOq/ZeRX14LAPBzs0fsg/0weZAnF58jEgEDCxFRK2obdPh3Ug5WH0hHaXU9gKbF516PCsS4wJ5cfI6oCxnz+7td90JXr14NX19fKJVKhIeHIzk5udW2X3/9NcLCwuDi4gJ7e3uEhITgs88+a9Zmzpw5kEgkzbbo6Oj2lEZEdEdKWxnmjvLDoUXjEftgPzgqbHA+vwIxm09g+rpEJGeWil0iEbXA6DssO3bswKxZs7B27VqEh4dj1apV+OKLL5CWlgZ3d/fb2ickJOD69esICgqCXC7Hd999h1dffRV79uxBVFQUgKbAUlBQgE2bNhmOUygUcHV1bVNNvMNCRO11vboeaw5mYMuxLNQ1Ni0+Ny6wJ16bGIj7vJ1Fro7IsnXqI6Hw8HAMGzYMn3zyCQBAr9dDrVbjpZdewpIlS9rUx9ChQzF58mSsWLECQFNgKSsrw65du4wpxYCBhYjulba8Fh/9eBk7T+SiUd/01+LkwZ549cF+8O/pIHJ1RJap0x4J1dfXIyUlBZGRkb90IJUiMjISiYmJdz1eEARoNBqkpaVhzJgxzT5LSEiAu7s7AgMDsWDBApSUlLTaT11dHSoqKpptRET3wsNZib8+Ogj7Y8diaogXJBJgz5l8PPj3Q1j85RlcK7shdolEVs2owFJcXAydTgeVStVsv0qlglarbfW48vJyODg4QC6XY/Lkyfj444/x4IMPGj6Pjo7G1q1bodFo8N577+HgwYOYNGkSdDpdi/3FxcXB2dnZsKnVamNOg4ioVb5u9vjHzCHY+8fRiOzvDp1ewI6TuRj3fgLe/u95lFTViV0ikVUy6pHQtWvX4O3tjWPHjiEiIsKwf9GiRTh48CCSkpJaPE6v1+PKlSuoqqqCRqPBihUrsGvXLowbN67F9leuXEFAQAD279+PCRMm3PZ5XV0d6up++UujoqICarWaj4SIqMOlZJdiZXwakm4OxrWXy/DcKD88N9IPrlzun+ieGPNIyKg1qt3c3CCTyVBQUNBsf0FBATw8PFo9TiqVok+fPgCAkJAQXLhwAXFxca0GFn9/f7i5uSE9Pb3FwKJQKKBQKIwpnYioXUJ9uuPz+ffj8OVivL8vDWevluPjH9Ox4Ugmfje8N54f7Q8PZ6XYZRJZPKMeCcnlcoSGhkKj0Rj26fV6aDSaZndc7kav1ze7Q/JbeXl5KCkpgaenpzHlERF1ColEgjH9emL3iyOx9pmhGOjlhJp6Hf51JBNjVh7A0q/PIOvmKrpE1DmMfgtYbGwsZs+ejbCwMAwfPhyrVq1CdXU1YmJiAACzZs2Ct7c34uLiADSNNwkLC0NAQADq6uqwd+9efPbZZ1izZg0AoKqqCm+99RYef/xxeHh4ICMjA4sWLUKfPn0M056JiEyBRCJB9H2eiBrogUOXi7H6QDqSM0uxPTkXO07k4uHBXlgwLgD9PflomqijGR1YZsyYgaKiIixbtgxarRYhISGIj483DMTNycmBVPrLjZvq6mq88MILyMvLQ7du3RAUFIRt27ZhxowZAACZTIYzZ85gy5YtKCsrg5eXFyZOnIgVK1bwsQ8RmSSJRIKx/XpibL+eOJFVik8PpONAWhF2/3QNu3+6hglB7nhhfB+E+rRtLSkiujsuzU9E1AHOXSvHpwkZ2Hs2H7f+Vr3fvzteGNcHo/u6ccl/ohbwXUJERCK5UlSFdQev4OtTeWjQNf31OsjbGQvHB2DiAA++ZJHoVxhYiIhEll9+A+sPZeI/ydmobWha8r+PuwMWjA3AlBAv2Mra9So3IovCwEJEZCJKquqw+VgWNh/LQmVtIwDA26Ubfj/WH9PD1FDaykSukEg8DCxERCamsrYB247nYMORTBTfXC3XzUGO50b54Zn7feCktBW5QqKux8BCRGSiaht0+OJkLtYevIKrN99P5Ki0wewIX8SM9EUPB86OJOvBwEJEZOIadHr896dr+DQhA+mFVQAApa0UM4f1xu/H+sPTuZvIFRJ1PgYWIiIzodcL+P58AT5NSMeZvHIAgFwmxe/Ce+OF8QFwd+Sy/2S5GFiIiMyMIAg4kl6Mj39sWj0XaLrjMjvCF78fG4DufNEiWSAGFiIiMyUIAo6ml+BvP6ThVE4ZgKY3RMeM9MO80f5wtuPgXLIcDCxERGZOEAQkpBXhg+/TcO5aBYCmwbnzRvsjZqQvHDmriCwAAwsRkYUQBAH7zhXg7z9cQlpBJQDA1c4Wvx8bgFkRPrCTG/1KOCKTwcBCRGRh9HoB353Nx6r9l3ClqBoA4OagwAvjAvC78N5cgI7MEgMLEZGFatTpsev0NfxDcwm5pU3ruHg4KfHiA30wPUwNuQ2X/CfzwcBCRGThGnR6fHEyDx//eBn55bUAgF6u3fDHCX3x2BBv2PBdRWQGGFiIiKxEbYMOnyfnYHVCBooqm5b893OzxyuRffHwYC/I+HZoMmEMLEREVuZGvQ6fHc/C2oNXUFpdDwDo6+6A2Af7IWqgB6QMLmSCGFiIiKxUVV0jthzLwrqDGai4+XboAZ5OiH2wHyb0d4dEwuBCpoOBhYjIypXfaMCGI5nYeCQTVXVNwSVY7YJXH+yH0X3dGFzIJDCwEBERAOB6dT3WHbqCLceycKNBBwAY5O2MmJG+eHiwF2cVkagYWIiIqJmiyjqsScjAv5OyUdeoBwD0dFTg2ft98Lvw3nBzUIhcIVkjBhYiImpRaXU9tifnYGtiFgoqmmYVyW2kmBrshZiRfhjgxb9DqeswsBAR0R016PTYezYfG49m4afcMsP+CP8eiBnpiwn9VZwSTZ2OgYWIiNosNec6Nh7JxP9+1kKnb/qV0Lu7HeaM8MWTYb34okXqNAwsRERktGtlN7A1MRvbk3NQfqMBAOCgsMGTYb0wZ4QvfHrYi1whWRoGFiIiarea+kZ8c+oqNh3NQnphFQBAIgEi+6sQM9IXEf49OC2aOgQDCxER3TNBEHD4cjE2Hs1EQlqRYX+QhyOeG+WHKcFefEs03RMGFiIi6lDphVXYciwLX6bkGdZz6WEvx9PhvfHM/T5wd1KKXCGZIwYWIiLqFOU1Dfj8RA62HMvCtZtvibaVSfDwYC88N9IPg3o5i1whmRMGFiIi6lSNOj2+P1+AjUcycTL7umH/MF9XzB8TgAlB7nzhIt0VAwsREXWZM3ll2HQ0C9+duYYGXdOvlD7uDpg/xh/TQry5/D+1ioGFiIi6XEFFLTYfy8K2xGxU3nzhospJgbmj/PDU8N5cz4Vuw8BCRESiqaxtwPbkHGw4kmlY/t9RaYNn7vdBzEhfuDtygC41YWAhIiLR1TXq8O3pa1h3MAMZRdUAALlMisdDvTFvtD/8ezqIXCGJjYGFiIhMhl4vQHOxEGsPZiDl5gBdiQSIGuCBP4wLQIjaRdwCSTQMLEREZJJOZpVi7cEM7L9QaNgX7tcdfxgXgHH9enIFXStjzO/vdg3dXr16NXx9faFUKhEeHo7k5ORW23799dcICwuDi4sL7O3tERISgs8++6xZG0EQsGzZMnh6eqJbt26IjIzE5cuX21MaERGZsDDf7vjX7GH44U9j8ERoL9jKJEjKLEXMphOY9I/D+Do1Dw06vdhlkgkyOrDs2LEDsbGxWL58OVJTUxEcHIyoqCgUFha22L579+7485//jMTERJw5cwYxMTGIiYnBvn37DG1WrlyJjz76CGvXrkVSUhLs7e0RFRWF2tra9p8ZERGZrL4qR3zwZDAOLRqPeaP9YC+X4aK2ErE7f8LYlQew4Ugmqm/ONCIC2vFIKDw8HMOGDcMnn3wCANDr9VCr1XjppZewZMmSNvUxdOhQTJ48GStWrIAgCPDy8sKrr76K1157DQBQXl4OlUqFzZs3Y+bMmXftj4+EiIjMW3lNA7YlZWPT0SwUVzXNLHLuZotZET6YPcIXbg4KkSukztBpj4Tq6+uRkpKCyMjIXzqQShEZGYnExMS7Hi8IAjQaDdLS0jBmzBgAQGZmJrRabbM+nZ2dER4e3mqfdXV1qKioaLYREZH5crazxcLxfXBk8Xj89dFB8O1hh/IbDfj4x3SMfPdH/GXXWWSXVItdJonIqMBSXFwMnU4HlUrVbL9KpYJWq231uPLycjg4OEAul2Py5Mn4+OOP8eCDDwKA4Thj+oyLi4Ozs7NhU6vVxpwGERGZKKWtDL8L7w3Nq+Ow5umhCO7ljLpGPbYdz8H4DxKw8D+p+Cm3TOwySQQ2XfGHODo64vTp06iqqoJGo0FsbCz8/f0xbty4dvW3dOlSxMbGGn6uqKhgaCEisiAyqQSTBnki+j4PHL/SNLPo4KUi7DmTjz1n8jHcrzvmj/bHA3xnkdUwKrC4ublBJpOhoKCg2f6CggJ4eHi0epxUKkWfPn0AACEhIbhw4QLi4uIwbtw4w3EFBQXw9PRs1mdISEiL/SkUCigUfJ5JRGTpJBIJIgJ6ICKgBy7kV2D94SvYffoakjNLkZxZCv+e9nh+lD8eG+oNpa1M7HKpExn1SEgulyM0NBQajcawT6/XQ6PRICIios396PV61NU1Dary8/ODh4dHsz4rKiqQlJRkVJ9ERGTZ+ns64cPpITi8eDx+P9YfjkobXCmqxv99cxYj3/0R/9h/GaXV9WKXSZ3E6EdCsbGxmD17NsLCwjB8+HCsWrUK1dXViImJAQDMmjUL3t7eiIuLA9A03iQsLAwBAQGoq6vD3r178dlnn2HNmjUAmtLzK6+8gnfeeQd9+/aFn58f3njjDXh5eWHatGkdd6ZERGQRPJ27Yemk/njpgb7YcSIXG49k4mrZDfx9/yV8mpCOJ0J7Ye4oPy79b2GMDiwzZsxAUVERli1bBq1Wi5CQEMTHxxsGzebk5EAq/eXGTXV1NV544QXk5eWhW7duCAoKwrZt2zBjxgxDm0WLFqG6uhrz589HWVkZRo0ahfj4eCiVfEEWERG1zEFhg7mj/DA7wgf/+1mLfx66grNXy/HvpBz8JzkHD/ZXYf4Yf4T6uHIFXQvApfmJiMgiCIKApMxSrD90BZqLvyxmGqJ2wfwx/oga6AEZB+iaFL5LiIiIrFp6YSU2HMnEV6lXUd/YtNR/7+52mDvKD0+G9YKdvEsmydJdMLAQEREBKKqsw2eJWdh6PBtlNQ0AmlbQfeb+3pgd4Qt3Jw49EBMDCxER0a/cqNfhy9Q8bDh8BVklNQAAuUyKaUO88Pxof/RTOYpcoXViYCEiImqBTi/gh/MFWH/4ClKyrxv2jwvsifmj/RER0IMDdLsQAwsREdFdpGRfx78OX0H8OS1u/Sbsp3LA9DA1pg3x5gsXuwADCxERURtlFVdj49FM7DyZi9qGpgG6NlIJIvurMH1YL4zp2xM2MqPWWaU2YmAhIiIyUkVtA/770zXsPJGLn/LKDftVTgo8PrQXngxTw8/NXsQKLQ8DCxER0T24qK3AFyfz8M2pq82W+x/u1x3Tw9R4aJAHp0Z3AAYWIiKiDlDfqIfmQgF2nMzFoUtF0N/8jemgsMEjwZ54MkyNIWoXDtRtJwYWIiKiDpZffgNfp17FzpO5yL45NRoA+ro3DdR9dCgH6hqLgYWIiKiT6PUCkrNKsfNELvb+nN9soO6E/u6YHqbG2H4cqNsWDCxERERdoKK2Ad/9lI8dJ3PxU26ZYb+7owKPh/bCdA7UvSMGFiIioi6Wpq3EzpO5tw/U9e2OJ8N6YfJgTw7U/Q0GFiIiIpHUN+rx48UC7DiRi4O/GqhrL5dhTL+eGB/ojrGBPaHie4wYWIiIiEyBtrwWX6Xm3TZQFwAGejlhfKA7xgf1RIjaFTKp9c00YmAhIiIyIYIg4Ke8ciSkFeJAWhHO5JXh1799XexsMaZvT4wP6okxfXuih5XMNmJgISIiMmHFVXU4dKkIB9KKcOhSEcpvNBg+k0iAwb1cMD6w6fHRIG9nSC307gsDCxERkZlo1OlxOrcMB9IKceBiEc7nVzT7vIe9HGNvhpcxfXvC2c5WpEo7HgMLERGRmdKW1+LgpabwciS9GFV1jYbPpBIg1McV4wLdMT7QHf09Hc16lV0GFiIiIgtQ36jHyexSJKQV4cDFQlwurGr2ucpJgfGB7hgX6I5Rfd3goDCvadMMLERERBYot7QGCZeKkHCxEEczig2r7AKA0laKx4b2wnMj/dDH3UHEKtuOgYWIiMjC1TbokJRZigMXC3EgrbDZtOnxgT0xd5Q/RvbpYdKPjBhYiIiIrIggCEjKLMW/DmdCc7HAMGU6yMMRz43yw9QQLyhsZOIW2QIGFiIiIiuVWVyNzUczsfNkHm406AAAbg5yPHu/L56+v7dJvVGagYWIiMjKldc0YPuJHGw5loX88loAgNxGikdDvDF3tB/6qRxFrpCBRexyiIiITEaDTo///azFhsNX8FNeuWH/6L5umDvKD2P79RRtnAsDCxERETUjCAJSsq/jX4cz8f15reGljH3dHfDcKD88OsQbStuuHefCwEJEREStyi2twaajWdh5MtewMF13ezmeDu+NZyN84O7YNW+SZmAhIiKiu6qobcDOE7nYdDQLV8tuAADkMikeCfbC3FF+GODVub9TGViIiIiozRp1enx/vgD/OnwFqTllhv0jAnpg7ig/jA9075QXMDKwEBERUbucyrmODUcy8b+ftdDdHOji72aPmJG+eCJUjW7yjhvnwsBCRERE9+Rq2Q1sOZaF7ck5qKxthNxGimNLHujQdVyM+f1tXm9JIiIioi7h7dIN//dQf/xxQl98cTIX12saRF10joGFiIiIWuWgsEHMSD+xy4BU7AKIiIiI7qZdgWX16tXw9fWFUqlEeHg4kpOTW227fv16jB49Gq6urnB1dUVkZORt7efMmQOJRNJsi46Obk9pREREZIGMDiw7duxAbGwsli9fjtTUVAQHByMqKgqFhYUttk9ISMBTTz2FAwcOIDExEWq1GhMnTsTVq1ebtYuOjkZ+fr5h2759e/vOiIiIiCyO0bOEwsPDMWzYMHzyyScAAL1eD7VajZdeeglLliy56/E6nQ6urq745JNPMGvWLABNd1jKysqwa9cu488AnCVERERkjoz5/W3UHZb6+nqkpKQgMjLylw6kUkRGRiIxMbFNfdTU1KChoQHdu3dvtj8hIQHu7u4IDAzEggULUFJS0mofdXV1qKioaLYRERGR5TIqsBQXF0On00GlUjXbr1KpoNVq29TH4sWL4eXl1Sz0REdHY+vWrdBoNHjvvfdw8OBBTJo0CTqdrsU+4uLi4OzsbNjUarUxp0FERERmpkunNb/77rv4/PPPkZCQAKXylxcrzZw50/DPgwYNwuDBgxEQEICEhARMmDDhtn6WLl2K2NhYw88VFRUMLURERBbMqDssbm5ukMlkKCgoaLa/oKAAHh4edzz2gw8+wLvvvovvv/8egwcPvmNbf39/uLm5IT09vcXPFQoFnJycmm1ERERkuYwKLHK5HKGhodBoNIZ9er0eGo0GERERrR63cuVKrFixAvHx8QgLC7vrn5OXl4eSkhJ4enoaUx4RERFZKKOnNcfGxmL9+vXYsmULLly4gAULFqC6uhoxMTEAgFmzZmHp0qWG9u+99x7eeOMNbNy4Eb6+vtBqtdBqtaiqqgIAVFVV4fXXX8fx48eRlZUFjUaDqVOnok+fPoiKiuqg0yQiIiJzZvQYlhkzZqCoqAjLli2DVqtFSEgI4uPjDQNxc3JyIJX+koPWrFmD+vp6PPHEE836Wb58Od58803IZDKcOXMGW7ZsQVlZGby8vDBx4kSsWLECCoV47ywgIiIi08G3NRMREZEoOm0dFiIiIiIxWMTbmm/dJOICckRERObj1u/ttjzssYjAUllZCQBci4WIiMgMVVZWwtnZ+Y5tLGIMi16vx7Vr1+Do6AiJRNKhfd9alC43N9fix8dY07kC1nW+PFfLZU3ny3O1PIIgoLKyEl5eXs0m7LTEIu6wSKVS9OrVq1P/DGtaoM6azhWwrvPluVouazpfnqtlududlVs46JaIiIhMHgMLERERmTwGlrtQKBRYvny5VSxiZ03nCljX+fJcLZc1nS/P1bpZxKBbIiIismy8w0JEREQmj4GFiIiITB4DCxEREZk8BhYiIiIyeQwsAFavXg1fX18olUqEh4cjOTn5ju2/+OILBAUFQalUYtCgQdi7d28XVdp+cXFxGDZsGBwdHeHu7o5p06YhLS3tjsds3rwZEomk2aZUKruo4nvz5ptv3lZ7UFDQHY8xx+sKAL6+vredq0QiwcKFC1tsb27X9dChQ3jkkUfg5eUFiUSCXbt2NftcEAQsW7YMnp6e6NatGyIjI3H58uW79mvs974r3OlcGxoasHjxYgwaNAj29vbw8vLCrFmzcO3atTv22Z7vQle423WdM2fObXVHR0fftV9TvK7A3c+3pe+wRCLB+++/32qfpnptO4vVB5YdO3YgNjYWy5cvR2pqKoKDgxEVFYXCwsIW2x87dgxPPfUU5s6di1OnTmHatGmYNm0afv755y6u3DgHDx7EwoULcfz4cfzwww9oaGjAxIkTUV1dfcfjnJyckJ+fb9iys7O7qOJ7N3DgwGa1HzlypNW25npdAeDEiRPNzvOHH34AADz55JOtHmNO17W6uhrBwcFYvXp1i5+vXLkSH330EdauXYukpCTY29sjKioKtbW1rfZp7Pe+q9zpXGtqapCamoo33ngDqamp+Prrr5GWloYpU6bctV9jvgtd5W7XFQCio6Ob1b19+/Y79mmq1xW4+/n++jzz8/OxceNGSCQSPP7443fs1xSvbacRrNzw4cOFhQsXGn7W6XSCl5eXEBcX12L76dOnC5MnT262Lzw8XPj973/fqXV2tMLCQgGAcPDgwVbbbNq0SXB2du66ojrQ8uXLheDg4Da3t5TrKgiC8PLLLwsBAQGCXq9v8XNzvq4AhG+++cbws16vFzw8PIT333/fsK+srExQKBTC9u3bW+3H2O+9GH57ri1JTk4WAAjZ2dmttjH2uyCGls519uzZwtSpU43qxxyuqyC07dpOnTpVeOCBB+7YxhyubUey6jss9fX1SElJQWRkpGGfVCpFZGQkEhMTWzwmMTGxWXsAiIqKarW9qSovLwcAdO/e/Y7tqqqq4OPjA7VajalTp+LcuXNdUV6HuHz5Mry8vODv74+nn34aOTk5rba1lOtaX1+Pbdu24bnnnrvji0DN+br+WmZmJrRabbNr5+zsjPDw8FavXXu+96aqvLwcEokELi4ud2xnzHfBlCQkJMDd3R2BgYFYsGABSkpKWm1rSde1oKAAe/bswdy5c+/a1lyvbXtYdWApLi6GTqeDSqVqtl+lUkGr1bZ4jFarNaq9KdLr9XjllVcwcuRI3Hfffa22CwwMxMaNG/Htt99i27Zt0Ov1GDFiBPLy8rqw2vYJDw/H5s2bER8fjzVr1iAzMxOjR49GZWVli+0t4boCwK5du1BWVoY5c+a02sacr+tv3bo+xly79nzvTVFtbS0WL16Mp5566o4vxzP2u2AqoqOjsXXrVmg0Grz33ns4ePAgJk2aBJ1O12J7S7muALBlyxY4Ojriscceu2M7c7227WURb2sm4yxcuBA///zzXZ91RkREICIiwvDziBEj0L9/f6xbtw4rVqzo7DLvyaRJkwz/PHjwYISHh8PHxwc7d+5s0/+1mKsNGzZg0qRJ8PLyarWNOV9XatLQ0IDp06dDEASsWbPmjm3N9bswc+ZMwz8PGjQIgwcPRkBAABISEjBhwgQRK+t8GzduxNNPP33XwfDmem3by6rvsLi5uUEmk6GgoKDZ/oKCAnh4eLR4jIeHh1HtTc2LL76I7777DgcOHECvXr2MOtbW1hZDhgxBenp6J1XXeVxcXNCvX79Wazf36woA2dnZ2L9/P55//nmjjjPn63rr+hhz7drzvTclt8JKdnY2fvjhhzveXWnJ3b4Lpsrf3x9ubm6t1m3u1/WWw4cPIy0tzejvMWC+17atrDqwyOVyhIaGQqPRGPbp9XpoNJpm/wf6axEREc3aA8APP/zQantTIQgCXnzxRXzzzTf48ccf4efnZ3QfOp0OZ8+ehaenZydU2LmqqqqQkZHRau3mel1/bdOmTXB3d8fkyZONOs6cr6ufnx88PDyaXbuKigokJSW1eu3a8703FbfCyuXLl7F//3706NHD6D7u9l0wVXl5eSgpKWm1bnO+rr+2YcMGhIaGIjg42OhjzfXatpnYo37F9vnnnwsKhULYvHmzcP78eWH+/PmCi4uLoNVqBUEQhGeffVZYsmSJof3Ro0cFGxsb4YMPPhAuXLggLF++XLC1tRXOnj0r1im0yYIFCwRnZ2chISFByM/PN2w1NTWGNr8917feekvYt2+fkJGRIaSkpAgzZ84UlEqlcO7cOTFOwSivvvqqkJCQIGRmZgpHjx4VIiMjBTc3N6GwsFAQBMu5rrfodDqhd+/ewuLFi2/7zNyva2VlpXDq1Cnh1KlTAgDhww8/FE6dOmWYGfPuu+8KLi4uwrfffiucOXNGmDp1quDn5yfcuHHD0McDDzwgfPzxx4af7/a9F8udzrW+vl6YMmWK0KtXL+H06dPNvsd1dXWGPn57rnf7LojlTudaWVkpvPbaa0JiYqKQmZkp7N+/Xxg6dKjQt29foba21tCHuVxXQbj7f8eCIAjl5eWCnZ2dsGbNmhb7MJdr21msPrAIgiB8/PHHQu/evQW5XC4MHz5cOH78uOGzsWPHCrNnz27WfufOnUK/fv0EuVwuDBw4UNizZ08XV2w8AC1umzZtMrT57bm+8sorhn8vKpVKeOihh4TU1NSuL74dZsyYIXh6egpyuVzw9vYWZsyYIaSnpxs+t5Tresu+ffsEAEJaWtptn5n7dT1w4ECL/+3eOie9Xi+88cYbgkqlEhQKhTBhwoTb/j34+PgIy5cvb7bvTt97sdzpXDMzM1v9Hh84cMDQx2/P9W7fBbHc6VxramqEiRMnCj179hRsbW0FHx8fYd68ebcFD3O5roJw9/+OBUEQ1q1bJ3Tr1k0oKytrsQ9zubadRSIIgtCpt3CIiIiI7pFVj2EhIiIi88DAQkRERCaPgYWIiIhMHgMLERERmTwGFiIiIjJ5DCxERERk8hhYiIiIyOQxsBAREZHJY2AhIiIik8fAQkRERCaPgYWIiIhMHgMLERERmbz/D3hr+MKhWUMQAAAAAElFTkSuQmCC\n"
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "## Save LoRA for each epoch"
      ],
      "metadata": {
        "id": "Q_M1J-ewJm--"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "for x in range(20):\n",
        "  gemma.fit(train, epochs=1, batch_size=2)\n",
        "  model_name = f\"/content/drive/MyDrive/{lora_name}_{lora_rank}_epoch{x+1}.lora.h5\"\n",
        "  gemma.backbone.save_lora_weights(model_name)\n",
        "  text_gen(\"나랏말ᄊᆞ미\")\n",
        "  text_gen(\"나는월궁의션녀로셔\")\n",
        "  text_gen(\"일쟝춘몽이라\")\n",
        "  text_gen(\"션녀왈\")\n"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "-27fD7pUJaxO",
        "outputId": "84570d28-1045-403b-ccd5-2a1e59e9c986"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\u001b[1m224/224\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m121s\u001b[0m 311ms/step - loss: 1.2034 - sparse_categorical_accuracy: 0.2637\n",
            "\n",
            "Gemma output:\n",
            "나랏말ᄊᆞ미ᄯᆞ여졔졀이라\n",
            "TOTAL TIME ELAPSED: 15.35s\n",
            "\n",
            "Gemma output:\n",
            "나는월궁의션녀로셔졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔이졔\n",
            "TOTAL TIME ELAPSED: 2.50s\n",
            "\n",
            "Gemma output:\n",
            "일쟝춘몽이라졔졔이졔졔이졔졔이졔졔이졔졔이졔졔이졔졔이졔졔이졔졔이졔졔이졔졔이졔졔이졔졔이졔졔이졔졔이졔졔이졔졔이졔졔이졔졔이졔졔이졔졔이졔졔이졔졔이졔졔이졔졔이졔졔이졔졔이졔졔이졔졔이졔졔이졔졔이졔졔이졔졔이졔졔이졔졔이졔�\n",
            "TOTAL TIME ELAPSED: 2.52s\n",
            "\n",
            "Gemma output:\n",
            "션녀왈쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔\n",
            "TOTAL TIME ELAPSED: 2.55s\n",
            "\u001b[1m224/224\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m52s\u001b[0m 232ms/step - loss: 0.9883 - sparse_categorical_accuracy: 0.3212\n",
            "\n",
            "Gemma output:\n",
            "나랏말ᄊᆞ미ᄒᆞᆫ지라\n",
            "TOTAL TIME ELAPSED: 0.09s\n",
            "\n",
            "Gemma output:\n",
            "나는월궁의션녀로셔ᄒᆞᆫ지라\n",
            "TOTAL TIME ELAPSED: 0.09s\n",
            "\n",
            "Gemma output:\n",
            "일쟝춘몽이라ᄒᆞᆫ지라\n",
            "TOTAL TIME ELAPSED: 0.10s\n",
            "\n",
            "Gemma output:\n",
            "션녀왈쳔쳔의쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔쳔��\n",
            "TOTAL TIME ELAPSED: 2.55s\n",
            "\u001b[1m224/224\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m52s\u001b[0m 232ms/step - loss: 0.9248 - sparse_categorical_accuracy: 0.3544\n",
            "\n",
            "Gemma output:\n",
            "나랏말ᄊᆞ미의ᄉᆞᆫᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯ\n",
            "TOTAL TIME ELAPSED: 2.52s\n",
            "\n",
            "Gemma output:\n",
            "나는월궁의션녀로셔ᄒᆞᆫ쳔졀의졀쳔을셩ᄒᆞ야셩셩을셩ᄒᆞ야셩셩을셩ᄒᆞ고셩셩을셩ᄒᆞ고셩셩을셩ᄒᆞ고셩셩을셩ᄒᆞ고셩셩을셩ᄒᆞ고셩셩을셩ᄒᆞ고셩셩을셩ᄒᆞ고셩셩을셩ᄒᆞ고셩셩을셩ᄒᆞ고셩셩을셩ᄒᆞ고셩셩을셩ᄒᆞ고셩셩을셩ᄒᆞ고셩셩을셩ᄒᆞ고셩셩을셩ᄒᆞ고셩셩을셩ᄒᆞ고셩셩을셩ᄒᆞ고셩\n",
            "TOTAL TIME ELAPSED: 2.50s\n",
            "\n",
            "Gemma output:\n",
            "일쟝춘몽이라ᄒᆞᆫ지라\n",
            "TOTAL TIME ELAPSED: 0.09s\n",
            "\n",
            "Gemma output:\n",
            "션녀왈쳔젹이젹의젹을젹ᄒᆞ고젹을젹ᄒᆞ고젹을젹ᄒᆞ고젹을젹ᄒᆞ고젹을젹ᄒᆞ고젹을젹ᄒᆞ고젹을젹ᄒᆞ고젹을젹ᄒᆞ고젹을젹ᄒᆞ고젹을젹ᄒᆞ고젹을젹ᄒᆞ고젹을젹ᄒᆞ고젹을젹ᄒᆞ고젹을젹ᄒᆞ고젹을젹ᄒᆞ고젹을젹ᄒᆞ고젹을젹ᄒᆞ고젹을젹ᄒᆞ고젹을젹ᄒᆞ고젹을젹ᄒᆞ고젹을젹ᄒᆞ고젹을젹ᄒᆞ고젹을젹ᄒᆞ고젹을젹ᄒᆞ고�\n",
            "TOTAL TIME ELAPSED: 2.55s\n",
            "\u001b[1m224/224\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m52s\u001b[0m 232ms/step - loss: 0.8761 - sparse_categorical_accuracy: 0.3820\n",
            "\n",
            "Gemma output:\n",
            "나랏말ᄊᆞ미의ᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅᆞᆫᄉᆞᄅ\n",
            "TOTAL TIME ELAPSED: 2.52s\n",
            "\n",
            "Gemma output:\n",
            "나는월궁의션녀로셔ᄒᆞᆫ지라\n",
            "TOTAL TIME ELAPSED: 0.09s\n",
            "\n",
            "Gemma output:\n",
            "일쟝춘몽이라ᄒᆞᆫᄉᆞᄅᆞᆷ이쳔졀의ᄒᆞᆫᄉᆞᄅᆞᆷ이라ᄒᆞ고\n",
            "TOTAL TIME ELAPSED: 0.34s\n",
            "\n",
            "Gemma output:\n",
            "션녀왈쳔젹이젹을젹ᄒᆞ고젹젹을젹ᄒᆞ고젹젹을젹ᄒᆞ고젹젹을젹ᄒᆞ고젹젹을젹ᄒᆞ고젹젹을젹ᄒᆞ고젹젹을젹ᄒᆞ고젹젹을젹ᄒᆞ고젹젹을젹ᄒᆞ고젹젹을젹ᄒᆞ고젹젹을젹ᄒᆞ고젹젹을젹ᄒᆞ고젹젹을젹ᄒᆞ고젹젹을젹ᄒᆞ고젹젹을젹ᄒᆞ고젹젹을젹ᄒᆞ고젹젹을젹ᄒᆞ고젹젹을젹ᄒᆞ고젹젹을젹ᄒᆞ고�\n",
            "TOTAL TIME ELAPSED: 2.55s\n",
            "\u001b[1m224/224\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m52s\u001b[0m 232ms/step - loss: 0.8326 - sparse_categorical_accuracy: 0.4045\n",
            "\n",
            "Gemma output:\n",
            "나랏말ᄊᆞ미ᄉᆞᆫᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄉᆞᄅᆞᆫᄃᆡᄉᆞᄅᆞᆫᄃᆡᄉᆞᄅᆞᆫᄃᆡᄉᆞᄅᆞᆫᄃᆡ\n",
            "TOTAL TIME ELAPSED: 0.53s\n",
            "\n",
            "Gemma output:\n",
            "나는월궁의션녀로셔졔의졀을셩ᄒᆞ니ᄉᆞᄅᆞᆷ이쳔셩을셩ᄒᆞ고졔졀을셩ᄒᆞ고졔졀을셩ᄒᆞ고졔졀을셩ᄒᆞ고졔졀을셩ᄒᆞ고졔졀을셩ᄒᆞ고졔졀을셩ᄒᆞ고졔졀을셩ᄒᆞ고졔졀을셩ᄒᆞ고졔졀을셩ᄒᆞ고졔졀을셩ᄒᆞ고졔졀을셩ᄒᆞ고졔졀을셩ᄒᆞ고졔졀을셩ᄒᆞ고졔졀을셩ᄒᆞ고졔졀을셩ᄒᆞ고졔졀을셩ᄒᆞ고졔졀\n",
            "TOTAL TIME ELAPSED: 2.50s\n",
            "\n",
            "Gemma output:\n",
            "일쟝춘몽이라ᄒᆞᆫᄉᆞᄅᆞᆷ이라ᄒᆞ고\n",
            "TOTAL TIME ELAPSED: 0.17s\n",
            "\n",
            "Gemma output:\n",
            "션녀왈쳔왕이쳔왕을쳔ᄒᆞ고쳔왕을쳔ᄒᆞ고쳔왕을쳔ᄒᆞ고쳔왕을쳔ᄒᆞ고쳔왕을쳔ᄒᆞ고쳔왕을쳔ᄒᆞ고쳔왕을쳔ᄒᆞ고쳔왕을쳔ᄒᆞ고쳔왕을쳔ᄒᆞ고쳔왕을쳔ᄒᆞ고쳔왕을쳔ᄒᆞ고쳔왕을쳔ᄒᆞ고쳔왕을쳔ᄒᆞ고쳔왕을쳔ᄒᆞ고쳔왕을쳔ᄒᆞ고쳔왕을쳔ᄒᆞ고쳔왕을쳔ᄒᆞ고쳔왕을쳔ᄒᆞ고쳔왕을쳔ᄒᆞ고쳔왕을쳔ᄒᆞ고쳔왕을쳔ᄒᆞ고쳔왕을쳔ᄒᆞ고쳔왕을\n",
            "TOTAL TIME ELAPSED: 2.55s\n",
            "\u001b[1m224/224\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m52s\u001b[0m 232ms/step - loss: 0.7928 - sparse_categorical_accuracy: 0.4283\n",
            "\n",
            "Gemma output:\n",
            "나랏말ᄊᆞ미ᄭᅬᄒᆞ여왈그ᄃᆡᄒᆞᆫ졔ᄉᆞᆷ이졔장이라\n",
            "TOTAL TIME ELAPSED: 0.36s\n",
            "\n",
            "Gemma output:\n",
            "나는월궁의션녀로셔ᄒᆞᆫ지라\n",
            "TOTAL TIME ELAPSED: 0.09s\n",
            "\n",
            "Gemma output:\n",
            "일쟝춘몽이라ᄒᆞᆫ지라\n",
            "TOTAL TIME ELAPSED: 0.09s\n",
            "\n",
            "Gemma output:\n",
            "션녀왈쳔의ᄒᆞᆫ쳡장이쳔셩을셩ᄒᆞ야쳔셩을셩ᄒᆞ야쳔셩을ᄉᆞᄅᆞᆯᄉᆞᄅᆞᆯᄀᆞᆫᄃᆡ\n",
            "TOTAL TIME ELAPSED: 0.66s\n",
            "\u001b[1m224/224\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m52s\u001b[0m 232ms/step - loss: 0.7600 - sparse_categorical_accuracy: 0.4482\n",
            "\n",
            "Gemma output:\n",
            "나랏말ᄊᆞ미ᄭᅬᄒᆞ여왈그ᄃᆡᄒᆞᆫ졔ᄒᆞᆫᄃᆡ장이졀장이라\n",
            "TOTAL TIME ELAPSED: 0.40s\n",
            "\n",
            "Gemma output:\n",
            "나는월궁의션녀로셔션ᄒᆞᆫ고쳔의쳡장을ᄌᆞᄒᆞ고쳡장쳔쳡을ᄌᆞᄒᆞ고쳡장쳡을ᄌᆞᄒᆞ고션쳔을ᄌᆞᄒᆞ고쳡쳡을ᄌᆞᄒᆞ고쳡쳡을ᄌᆞᄒᆞ고션ᄉᆞᆷ을ᄌᆞᄒᆞ고쳡쳡쳡을ᄌᆞᄒᆞ고쳡쳡쳡을ᄌᆞᄒᆞ고쳡쳡을ᄌᆞᄒᆞ고션ᄃᆡ쳡을ᄌᆞᄒᆞ고쳡쳡을ᄌᆞᄒᆞ고쳡쳡을ᄌᆞᄒᆞ고션ᄃᆡ쳡을ᄌᆞᄒᆞ고쳡쳡쳡을ᄌᆞᄒᆞ고쳡쳡쳡을ᄌᆞᄒᆞ고션ᄃᆡ쳡을ᄌᆞᄒᆞ고쳡쳡을ᄌᆞᄒᆞ고쳡��\n",
            "TOTAL TIME ELAPSED: 2.50s\n",
            "\n",
            "Gemma output:\n",
            "일쟝춘몽이라ᄒᆞᆫ고졔졀의ᄒᆞᆫᄒᆞᆫᄒᆞᆫᄒᆞᆫᄒᆞᆫᄉᆞᄅᆞᆷ이라ᄒᆞ고\n",
            "TOTAL TIME ELAPSED: 0.40s\n",
            "\n",
            "Gemma output:\n",
            "션녀왈쳔왕이왈쳔왕이왈쳔왕이왈\n",
            "TOTAL TIME ELAPSED: 0.22s\n",
            "\u001b[1m224/224\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m52s\u001b[0m 232ms/step - loss: 0.7391 - sparse_categorical_accuracy: 0.4580\n",
            "\n",
            "Gemma output:\n",
            "나랏말ᄊᆞ미ᄒᆞ여왈필리졀의말왈너희ᄂᆞᆫ쳔ᄉᆡᆼ이ᄉᆞᆷ왈너희ᄂᆞᆫ쳔ᄉᆡᆼ이그ᄂᆞᆫ말을치고왈너희ᄂᆞᆫ말은너희쳔ᄉᆡᆼ의말이라\n",
            "TOTAL TIME ELAPSED: 0.78s\n",
            "\n",
            "Gemma output:\n",
            "나는월궁의션녀로셔션ᄒᆞᆫ고쳔통ᄉᆞᆫ쳘셩을ᄌᆞᆷᄒᆞ고쳘셩을ᄃᆡᄒᆞ야쳘셩을ᄌᆞᆷᄒᆞ고통ᄉᆞᆫ을ᄃᆡᄒᆞ야통ᄉᆞᆫ을ᄌᆡ조ᄒᆞ여통ᄉᆞᆫ을ᄌᆡᄒᆞ여통ᄉᆞᆫ을ᄌᆡᄒᆞ고통ᄉᆞᆫ을통ᄉᆞᆫᄒᆞ고통ᄉᆞᆫ을통ᄉᆞᆫᄒᆞ여통ᄉᆞᆫ을ᄌᆡᄒᆞ고통ᄉᆞᆫ을ᄃᆡᄒᆞ야통ᄉᆞᆫ을ᄌᆡᄒᆞ고통ᄉᆞᆫ을통ᄉᆞᆫᄒᆞ고통ᄉᆞᆫ을통ᄉᆞᆫᄒᆞ고통ᄉᆞᆫ을통ᄉᆞᆫᄒᆞ고통ᄉᆞᆫ을통ᄉᆞᆫᄒᆞ고통ᄉᆞᆫ을통ᄉᆞᆫᄒᆞ고통ᄉᆞᆫ을통ᄉᆞᆫᄒᆞ고통ᄉᆞᆫ을통ᄉᆞᆫᄒᆞ고통ᄉ\n",
            "TOTAL TIME ELAPSED: 2.50s\n",
            "\n",
            "Gemma output:\n",
            "일쟝춘몽이라엇지ᄒᆞᆫᄉᆞᄅᆞᆷ이ᄂᆡ여ᄒᆞᆫᄃᆡᄀᆞᆷ셩을모로고ᄒᆞᆫᄃᆡᄀᆞᆷ셩을엇지ᄒᆞ니\n",
            "TOTAL TIME ELAPSED: 0.56s\n",
            "\n",
            "Gemma output:\n",
            "션녀왈쳔왕이왈쳔왕의쳔왈쳔왕이왈쳔왕이왈\n",
            "TOTAL TIME ELAPSED: 0.31s\n",
            "\u001b[1m224/224\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m52s\u001b[0m 232ms/step - loss: 0.7188 - sparse_categorical_accuracy: 0.4696\n",
            "\n",
            "Gemma output:\n",
            "나랏말ᄊᆞ미ᄉᆞᆷ왈그मुᄒᆞᆫᄉᆞᄅᆞᆷ이ᄃᆡ장ᄒᆞ여ᄃᆡ진을ᄌᆞᆷᄒᆞ고ᄃᆡ장을ᄌᆞᆷᄒᆞ고ᄃᆡ장을ᄇᆡᆨᄒᆞ고ᄃᆡ장을ᄌᆞᆷᄒᆞ고ᄃᆡ장을ᄇᆡᆨᄒᆞ고ᄇᆡᆨ셩을ᄌᆞᆷᄒᆞ고ᄇᆡᆨ셩을ᄃᆡᄒᆞ야ᄇᆡᆨ셩을ᄌᆞᆷᄒᆞ고ᄇᆡᆨ셩을ᄃᆡᄒᆞ야ᄇᆡᆨ셩을ᄌᆞᆷᄒᆞ고ᄇᆡᆨ셩을ᄇᆡᆨ셩ᄒᆞ고ᄇᆡᆨ셩을ᄃᆡᄒᆞ야ᄇᆡᆨ셩을ᄌᆞᆷᄒᆞ고ᄇᆡᆨ셩을ᄇᆡᆨ셩ᄒᆞ고ᄇᆡᆨ셩을ᄃᆡᄒ\n",
            "TOTAL TIME ELAPSED: 2.52s\n",
            "\n",
            "Gemma output:\n",
            "나는월궁의션녀로셔션ᄒᆞᆫ고졀의ᄒᆞᆫ고졀이잇지못ᄒᆞᆯ고졀의ᄒᆞᆫ고졀이잇지못ᄒᆞᆯ고졀의ᄒᆞᆫ고졀이잇지못ᄒᆞ고고졈의ᄒᆞᆫ고졀이잇지못ᄒᆞ고\n",
            "TOTAL TIME ELAPSED: 0.84s\n",
            "\n",
            "Gemma output:\n",
            "일쟝춘몽이라엇지ᄒᆞᆫᄉᆞᄅᆞᆷ이ᄂᆡ여ᄒᆞᆫᄃᆡ글이잇잇잇ᄒᆞᆫᄃᆡ장ᄒᆞ야ᄉᆞᄅᆞᆷ을모로고일ᄀᆞᆫ장을엇지못ᄒᆞ야일ᄀᆞᆫ장을엇지못ᄒᆞ야일ᄀᆞᆫ장을엇지못ᄒᆞ야일ᄀᆞᆫ장을엇지못ᄒᆞ고일ᄯᆞ리라ᄒᆞ고\n",
            "TOTAL TIME ELAPSED: 1.10s\n",
            "\n",
            "Gemma output:\n",
            "션녀왈쳔의션ᄀᆞᆫ이ᄂᆡ의션ᄀᆞᆫ셩ᄉᆞᆫ셩ᄒᆞ고셩ᄉᆞᆫ셩ᄒᆞ고쳔셩ᄒᆞ고쳔셩ᄒᆞ고쳔ᄒᆞᆫ셩ᄒᆞ고쳔ᄭᅬᄒᆞᆫ셩ᄒᆞ고쳔ᄀᆞᆫ셩ᄀᆞᆫᄃᆡ\n",
            "TOTAL TIME ELAPSED: 1.10s\n",
            "\u001b[1m224/224\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m52s\u001b[0m 232ms/step - loss: 0.6981 - sparse_categorical_accuracy: 0.4815\n",
            "\n",
            "Gemma output:\n",
            "나랏말ᄊᆞ미ᄭᅮ지지고etcodeetcode\n",
            "TOTAL TIME ELAPSED: 0.14s\n",
            "\n",
            "Gemma output:\n",
            "나는월궁의션녀로셔션ᄒᆞᆫ고졀의ᄒᆞᆫ셩ᄒᆞᆫ셩ᄒᆞᆫᄉᆞᄅᆞᆷ이잇거니와졀졀의ᄉᆞᄅᆞᆷ이잇거니와고궁ᄒᆞᆫ고졔ᄂᆞᆫ지라\n",
            "TOTAL TIME ELAPSED: 0.69s\n",
            "\n",
            "Gemma output:\n",
            "일쟝춘몽이라ᄒᆞᆫ고졔졔ᄒᆞᆫ지라\n",
            "TOTAL TIME ELAPSED: 0.19s\n",
            "\n",
            "Gemma output:\n",
            "션녀왈젹젹의션젹을션젹ᄒᆞ시고졀셩을젹안의ᄃᆡᄒᆞᆫᄃᆡ젹이라ᄒᆞ시고졔셩을젹안ᄃᆡᄒᆞ시고ᄃᆡ감을ᄃᆡᄒᆞ시니ᄃᆡ감이ᄃᆡ안ᄒᆞ고ᄃᆡ감이젹안ᄒᆞ고ᄃᆡ감이젹안ᄒᆞ고ᄃᆡ감이ᄃᆡ안ᄒᆞ고ᄃᆡ감이젹안ᄒᆞ고ᄃᆡ감이ᄃᆡᄒᆞ시더라\n",
            "TOTAL TIME ELAPSED: 1.71s\n",
            "\u001b[1m224/224\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m52s\u001b[0m 232ms/step - loss: 0.6803 - sparse_categorical_accuracy: 0.4915\n",
            "\n",
            "Gemma output:\n",
            "나랏말ᄊᆞ미ᄯᆞ엇고지인의글ᄉᆞ로드리업셔위이라\n",
            "TOTAL TIME ELAPSED: 0.24s\n",
            "\n",
            "Gemma output:\n",
            "나는월궁의션녀로셔션듸오니ᄃᆡ감의ᄃᆡᄒᆞ야쳔명을ᄯᅥᄂᆞ오니ᄃᆡ감이쳔명을쳑분ᄒᆞ고ᄃᆡ감이ᄃᆡᄉᆞᄅᆞᆷ을ᄯᅥᄂᆞ오니ᄃᆡ감이ᄃᆡ감이ᄒᆞᆫ고졔명을쳑분ᄒᆞ오니ᄃᆡ감이ᄃᆡ감이ᄒᆞᆫ고졔명을ᄯᅥᄂᆞ오니ᄃᆡ감이ᄃᆡ감을ᄯᅥᄂᆞ오니ᄃᆡ감이ᄃᆡ감을ᄌᆞᆷ쳔ᄒᆞ고ᄃᆡ감이ᄃᆡ감을ᄌᆞᆷᄒᆞ고ᄃᆡ감이ᄒᆞᆫᄂᆡ셩쳔명을ᄌᆞᆷᄒᆞ고ᄃᆡ감이ᄃᆡ감을ᄌᆞᆷᄒᆞ고ᄃᆡ감이ᄃᆡ감\n",
            "TOTAL TIME ELAPSED: 2.50s\n",
            "\n",
            "Gemma output:\n",
            "일쟝춘몽이라ᄒᆞᆫ고졔ᄉᆡᆼ이라ᄒᆞ고\n",
            "TOTAL TIME ELAPSED: 0.21s\n",
            "\n",
            "Gemma output:\n",
            "션녀왈그졔ᄃᆡ왕의졔졀ᄒᆞ시더라\n",
            "TOTAL TIME ELAPSED: 0.25s\n",
            "\u001b[1m224/224\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m52s\u001b[0m 232ms/step - loss: 0.6629 - sparse_categorical_accuracy: 0.5019\n",
            "\n",
            "Gemma output:\n",
            "나랏말ᄊᆞ미셔왈그Dougall은오ᄃᆡᄉᆞᄅᆞᆷ을ᄌᆡ치고ᄃᆡ감을ᄌᆡᄐᆡᆨᄒᆞ니오ᄃᆞᄒᆞᆫᄉᆞᄅᆞᆷ이ᄉᆞᄅᆞᆷᄎᆞᆫ마음을ᄉᆞᆷᄂᆡ여ᄉᆞᄅᆞᆷ을ᄃᆡᄒᆞ야ᄉᆞᄅᆞᆷ마음을이르지못ᄒᆞ야ᄎᆞᆡ을알블ᄂᆡ여ᄉᆞᄅᆞᆷ마음을이ᄂᆡ여ᄉᆞᄅᆞᆷ마음을이르지아니ᄒᆞ리라ᄒᆞ고\n",
            "TOTAL TIME ELAPSED: 1.50s\n",
            "\n",
            "Gemma output:\n",
            "나는월궁의션녀로셔션롸ᄂᆞᆫᄃᆡ라ᄒᆞ야졔장이졀ᄌᆞᆷ셕을모로다ᄀᆞ즁ᄉᆞᆷᄉᆞᆫ이라\n",
            "TOTAL TIME ELAPSED: 0.50s\n",
            "\n",
            "Gemma output:\n",
            "일쟝춘몽이라ᄒᆞᆯᄉᆡ로ᄭᅳᆯᄒᆞᆫᄂᆡ을ᄒᆡᆼᄒᆞ고몽ᄒᆞᆫ몽ᄒᆞ고몽ᄒᆞᆫ몽ᄒᆞ고몽ᄒᆞᆫ몽ᄒᆞ고몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ몽ᄒᆞᆫ\n",
            "TOTAL TIME ELAPSED: 2.52s\n",
            "\n",
            "Gemma output:\n",
            "션녀왈그졔ᄉᆞᆫ이ᄂᆡᄒᆞᆫᄃᆡᄀᆞᆺᄐᆡᄒᆞ고ᄂᆡᄒᆞᆫᄃᆡᄀᆞᄒᆞᆫᄉᆞᄅᆞᆷ을ᄯᆞ라ᄒᆞ고\n",
            "TOTAL TIME ELAPSED: 0.63s\n",
            "\u001b[1m224/224\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m52s\u001b[0m 232ms/step - loss: 0.6426 - sparse_categorical_accuracy: 0.5164\n",
            "\n",
            "Gemma output:\n",
            "나랏말ᄊᆞ미ᄀᆞᆫᄉᆞᄅᆞᆷ을ᄌᆡموᄒᆞ야왈,ᄉᆞᆷ군의ᄌᆡ말을엇러ᄀᆞ지고왈,ᄂᆡᄒᆞᆫᄌᆡ말을엇지못ᄒᆞ야ᄉᆞᆷ군의ᄌᆡ말을알레니ᄒᆞ니,ᄃᆡ감이ᄒᆞᆫᄉᆞᄅᆞᆷ을ᄯᅥᄒᆞ더니,ᄃᆡ감이ᄉᆞᆷ군의ᄌᆡ말을알레니,ᄃᆡ감이ᄉᆞᆷ군을엇지ᄒᆞ야엇지ᄒᆞ고ᄒᆞᆫᄉᆞᄅᆞᆷ을알레더니,ᄃᆡ감이ᄉᆞᆷ군을알레더니,ᄃᆡ감이엇지못ᄒᆞ야ᄉᆞᆷ군을알레더니,ᄃᆡ감이엇지못ᄒᆞ야ᄉᆞᆷ군을알레더니,ᄃᆡ감이엇지못ᄒᆞ야ᄉᆞᆷ군을알레더니,ᄃᆡ감이엇지못ᄒᆞ야\n",
            "TOTAL TIME ELAPSED: 2.52s\n",
            "\n",
            "Gemma output:\n",
            "나는월궁의션녀로셔션롸ᄂᆞᆫ쳔ᄉᆡᆼ이오니ᄃᆡ감의ᄃᆡᆼᄒᆞᆯᄉᆡᆼ셩을모로더라\n",
            "TOTAL TIME ELAPSED: 0.49s\n",
            "\n",
            "Gemma output:\n",
            "일쟝춘몽이라ᄒᆞᆯᄉᆡ길동의ᄒᆞᆫᄉᆞᄅᆞᆷ을보ᄂᆡ여ᄉᆞᄅᆞᆷ을ᄒᆡᆼᄒᆞᄉᆞᆷᄒᆞ고ᄉᆞᄅᆞᆷ을ᄉᆞᆷᄒᆞ고ᄉᆡᆼᄒᆞᆫᄌᆡ조을ᄯᆞ라ᄒᆞ고\n",
            "TOTAL TIME ELAPSED: 0.82s\n",
            "\n",
            "Gemma output:\n",
            "션녀왈그쳔의ᄌᆞᆷᄉᆞᆷ을ᄌᆡ지처왈그ᄌᆡᄉᆞᆷ을모로고ᄌᆞᆷ을ᄌᆡ지처왈그ᄌᆡᄉᆞᆷ을ᄌᆡᄌ Jasa라그ᄌᆞᆷ을ᄌᆡᄌ Jasa라ᄒᆞ고ᄌᆡᄉᆞᆷ을ᄌᆡᄌ Jasa라ᄒᆞ고ᄌᆡᄉᆞᆷ을ᄌᆡᄌ Jasa라ᄒᆞ고그ᄌᆞᆷᄉᆞᆷ을모고그ᄌᆡᄉᆞᆷ을모와왈그ᄌᆡᄉᆞᆷ을모고ᄌᆡᄌᆡ지라ᄒᆞ고ᄌᆡᄌ Jasa라ᄒᆞ고그ᄌᆡᄉᆞᆷ을모와왈그ᄌᆡᄉᆞᆷ을모고ᄌᆡᄌ jurassic라\n",
            "TOTAL TIME ELAPSED: 1.99s\n",
            "\u001b[1m224/224\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m52s\u001b[0m 232ms/step - loss: 0.6095 - sparse_categorical_accuracy: 0.5363\n",
            "\n",
            "Gemma output:\n",
            "나랏말ᄊᆞ미ᄀᆞᆫᄉᆞᄅᆞᆷ을ᄌᆡموᄒᆞ여왈,ᄭᆡᄉᆞᆷ의블ᄉᆞ오ᄅᆞᆯvantaged이라ᄒᆞ니\n",
            "TOTAL TIME ELAPSED: 0.46s\n",
            "\n",
            "Gemma output:\n",
            "나는월궁의션녀로셔ᄃᆡ감왈ᄂᆡ감이엇더ᄒᆞᆫᄒᆞᆫᄃᆡ\n",
            "TOTAL TIME ELAPSED: 0.28s\n",
            "\n",
            "Gemma output:\n",
            "일쟝춘몽이라ᄒᆞᆯᄉᆡ길동의ᄒᆞᆫᄉᆞᄅᆞᆷ을보고ᄒᆞᆫᆫᄉᆡᆼ러ᄀᆞᆫᄉᆞᄅᆞᆷ을알셔ᄒᆞ시니길동이ᄉᆞᄅᆞᆷ을ᄌᆡ조ᄒᆞ여왈소여ᄅᆞᆯᄉᆞᄅᆞᆷ이라도ᄌᆡ조을ᄌᆞ길동ᄉᆞᆷ을ᄯᆞ라ᄒᆞ시니\n",
            "TOTAL TIME ELAPSED: 1.04s\n",
            "\n",
            "Gemma output:\n",
            "션녀왈우리션셤의션ᄒᆞᆫ쳑을ᄀᆞᆷ지못ᄒᆞ야ᄭᅳᆯ도와ᄒᆞᆫ소ᄅᆡ로션ᄀᆞᆫᄉᆞᄅᆞᆷ을보ᄂᆡ려ᄒᆞ고쳔쳡ᄒᆞᆫᄌᆡ조을ᄀᆞᆫᄃᆡᆨᄒᆞ시더라\n",
            "TOTAL TIME ELAPSED: 0.90s\n",
            "\u001b[1m224/224\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m52s\u001b[0m 233ms/step - loss: 0.5835 - sparse_categorical_accuracy: 0.5516\n",
            "\n",
            "Gemma output:\n",
            "나랏말ᄊᆞ미ᄀᆞᆫᄉᆞᄅᆞᆷ을ᄌᆡ dano ᄀᆞᄒᆡᆼᄒᆡᆼᄒᆞ더니\n",
            "TOTAL TIME ELAPSED: 0.35s\n",
            "\n",
            "Gemma output:\n",
            "나는월궁의션녀로셔ᄃᆡ감왈ᄂᆡ감슈아니ᄀᆞ라ᄒᆞᄂᆞᆫ지라\n",
            "TOTAL TIME ELAPSED: 0.28s\n",
            "\n",
            "Gemma output:\n",
            "일쟝춘몽이라ᄒᆞᆫᄉᆞᄅᆞᆷ이되여ᄀᆞᆷ슈ᄃᆡ이되니ᄃᆡ상ᄒᆞᆫᄉᆞᄅᆞᆷ이되고ᄃᆡ상ᄒᆞᆫᄉᆞᄅᆞᆷ이업고ᄉᆞᆫᄯᆞᆯᄉᆞᄅᆞᆷ이되니ᄃᆡ상ᄒᆞᆫᄉᆞᄅᆞᆷ이치고ᄃᆡ상ᄒᆞᆫᄉᆞᄅᆞᆷ이업고ᄉᆞᆫᄯᆞᆯᄉᆞᄅᆞᆷ이치고ᄒᆡᆼᄉᆡᆼᄅᆞᆯᄇᆡ업ᄂᆞᆫ지라\n",
            "TOTAL TIME ELAPSED: 1.45s\n",
            "\n",
            "Gemma output:\n",
            "션녀왈그ᄃᆡᄃᆡᄀᆞᆺᄐᆡᄉᆞᆷ미 ビᄒᆞ고ᄃᆡ감의ᄉᆞ엇더ᄒᆞᆫᄌᆡ조ᄃᆡ감이ᄒᆞᆫᄉᆞᄅᆞᆷ의직키고ᄒᆞᆫᄌᆡ조ᄃᆡ감이ᄒᆞ여ᄒᆞᆫᄌᆡ조ᄐᆡᄋᆞᆷᄒᆞ고ᄇᆡᆨ셩의ᄃᆡ감이ᄒᆞ여ᄒᆞᆫᄌᆡ조ᄐᆡᄋᆞᆷᄒᆞ고ᄃᆡ감의ᄇᆡᆨ셩의ᄌᆡᄉᆞᄒᆞᆫᄌᆡᄉᆞᄒᆞ고ᄌᆡ감의ᄇᆡᆨ셩의ᄌᆡᄉᆞᄒᆞ고ᄇᆡᆨ셩의ᄌᆡᄉᆞᄒᆞ고ᄌᆡ감의ᄇᆡᆨ셩의ᄌᆡᄉᆞᄒᆞ고ᄇᆡᆨ셩의ᄇᆡᆨ셩의ᄌᆡ\n",
            "TOTAL TIME ELAPSED: 2.55s\n",
            "\u001b[1m224/224\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m52s\u001b[0m 232ms/step - loss: 0.5618 - sparse_categorical_accuracy: 0.5663\n",
            "\n",
            "Gemma output:\n",
            "나랏말ᄊᆞ미ᄀᆞᆫᄉᆞᄅᆞᆯᄉᆞngrxrugated이십십블ᄒᆞᆫ말을엇 mię셔러ᄒᆞᆯ지니엇더ᄒᆞᆫᄉᆞᄅᆞᆷ을무ycznegoᄒᆞ고ᄉᆞᆷ의말을엇지ᄒᆞ니엇더ᄒᆞᆫᄉᆞᄅᆞᆷ을엇지ᄒᆞ니ᄒᆞᆫᄉᆞᄅᆞᆷ이ᄉᆞᆷ무ᄉᆞᆷᄒᆞ고ᄉᆞᆷ의말을엇지ᄒᆞ니엇더ᄒᆞᆫᄉᆞᄅᆞᆷ을엇지ᄒᆞ니ᄒᆞᆫᄉᆞᄅᆞᆷ이ᄉᆞᆷ무ᄉᆞᆷᄒᆞ고ᄉᆞᆷ의말을엇지ᄒᆞ니ᄒᆞᆫᄉᆞᄅᆞᆷ이ᄉᆞᆷ무ᄉᆞᆷᄒᆞ고ᄉᆞᆷ의말을엇엇지ᄒᆞ니ᄒᆞᆫᄉᆞᄅᆞᆷ이엇지ᄒᆞ고ᄉᆞᆷ의말을엇엇엇엇엇ᄒᆞ라ᄒᆞ고ᄒᆞᆫᄉᆞᄅᆞᆷ이엇지ᄒᆞ고ᄉᆞᆷ의말을엇엇엇엇엇엇ᄒᆞ라ᄒᆞ고ᄒᆞᆫᄉᆞᄅ\n",
            "TOTAL TIME ELAPSED: 2.52s\n",
            "\n",
            "Gemma output:\n",
            "나는월궁의션녀로셔션즁의ᄃᆡᄂᆞᆫᄃᆡ졔군을쳥ᄒᆞ여셩상졀의션즁을ᄌᆞᆷ고ᄉᆞ면을ᄌᆞᆷ고ᄃᆡ감을감ᄒᆞ야졔군을부러션즁의션쳥ᄒᆞ여셩상의션즁을ᄌᆞᆷ고ᄉᆞ면을ᄌᆞᆷ고ᄃᆡ감을감ᄒᆞ야졔군을부러쳠ᄉᆡᆼᄀᆞᄒᆞ거니와ᄉᆞ면을ᄌᆞᆷ고ᄃᆡ감을감 postale일ᄀᆞᄉᆞᆨ을ᄌᆞᆷ고ᄉᆞ면을ᄌᆞᆷ고감도부러ᄎᆞᆷᄌᆞ졔군을부러ᄒᆞ여졔군을알고ᄃᆡ감도부러ᄒᆞ여ᄎᆞᆷᄌᆞ졀셩을알고ᄃᆡᄎᆞᆷ도셩을알고ᄃᆡᄂᆞᆫ도셩ᄀ\n",
            "TOTAL TIME ELAPSED: 2.50s\n",
            "\n",
            "Gemma output:\n",
            "일쟝춘몽이라ᄃᆡᄀᆞᆷᄉᆞᆷᄒᆞ고ᄃᆡ감젼ᄒᆞ고ᄃᆡ감ᄉᆞᆷ이오니ᄉᆞᆷ쳔지ᄌᆞᆷᄉᆞ오ᄂᆞᆫᄃᆡ로ᄃᆡ감ᄉᆞᆷ이오ᄂᆞᆫ지라\n",
            "TOTAL TIME ELAPSED: 0.73s\n",
            "\n",
            "Gemma output:\n",
            "션녀왈그ᄃᆡᄃᆡᄀᆞᆺᄐᆡᄉᆞᆷ문을ᄯᅥᄂᆞ오니ᄉᆞᆷ문이허다ᄒᆞᆫ지라\n",
            "TOTAL TIME ELAPSED: 0.45s\n",
            "\u001b[1m224/224\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m52s\u001b[0m 232ms/step - loss: 0.5365 - sparse_categorical_accuracy: 0.5799\n",
            "\n",
            "Gemma output:\n",
            "나랏말ᄊᆞ미건지인이ᄉᆞᄅᆞᆷ을numerusform치지어ᄉᆞᄅᆞᆷ을 phù phù엇고rhosis이이ᄅᆞᆯᄲᅮᆫ여더라\n",
            "TOTAL TIME ELAPSED: 0.42s\n",
            "\n",
            "Gemma output:\n",
            "나는월궁의션녀로셔션즁의션러ᄀᆞ쳔ᄉᆡᆼ을보고ᄯᆞ라ᄒᆞ며ᄯᆞ여셩셩의ᄌᆞ veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste veste\n",
            "TOTAL TIME ELAPSED: 2.50s\n",
            "\n",
            "Gemma output:\n",
            "일쟝춘몽이라ᄃᆡᄀᆞᆷᄉᆞᆷᄒᆞ고ᄃᆡ감을감ᄉᆞᆷ이라ᄒᆞ고ᄃᆡ감ᄉᆞᆷ이라ᄒᆞ고ᄒᆞᆫᄆᆡ복ᄒᆞᆫᄉᆞᄅᆞᆷ을블너왈무ᄉᆞᆷ이잇스ᄂᆞᆫᄃᆡ장이ᄂᆡ미잇시면ᄉᆞᄅᆞᆷ을블너왈무ᄉᆞᆷ이잇스ᄂᆞᆫᄃᆡ장은군ᄉᆞ로ᄃᆡ장을모로다ᄒᆞ고ᄉᆞᄅᆞᆷ은ᄉᆞᄅᆞᆷ셩을모ᄒᆞ고ᄉᆞᄅᆞᆷ은ᄃᆡ감을모ᄒᆞ고ᄉᆞᄅᆞᆷ은ᄉᆞᄅᆞᆷ장을모ᄒᆞ라ᄒᆞ고\n",
            "TOTAL TIME ELAPSED: 1.85s\n",
            "\n",
            "Gemma output:\n",
            "션녀왈그ᄃᆡᄃᆡ왕의ᄉᆞᄅᆞᆯᄉᆞᄃᆡ을쳥ᄒᆞ여ᄉᆞᆷ셩을ᄌᆞᄇᆞ다ᄀᆞᆫᄉᆞᄅᆞᆯ좌왈그ᄃᆡᄃᆡ왕은ᄂᆡ의치로왈너희를모로더니너희를위ᄒᆞ여ᄉᆞᆷ셩을ᄌᆞᄇᆞ다ᄒᆞ여ᄉᆡᆼ각ᄒᆞ고도젹ᄒᆞ며ᄌᆞᄇᆞ다ᄒᆞ니그ᄃᆡᄃᆡ왕이ᄒᆞᆫᄂᆡ의치을쳔ᄒᆞ여도젹이와ᄒᆞᆫᄂᆡ의쳔명을ᄌᆡᄇᆞ오ᄂᆡᄂᆡ의드러ᄀᆞ도젹ᄒᆞ오ᄂᆡ다ᄒᆞ고셩상이그ᄃᆡ을부러ᄉᆞᆷ셩을ᄌᆞᄇᆞ도젹ᄒᆞ오ᄂᆡ다ᄒᆞ고도젹이와ᄒᆞᆫ\n",
            "TOTAL TIME ELAPSED: 2.55s\n",
            "\u001b[1m224/224\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m52s\u001b[0m 232ms/step - loss: 0.5161 - sparse_categorical_accuracy: 0.5942\n",
            "\n",
            "Gemma output:\n",
            "나랏말ᄊᆞ미건지의ᄒᆡᆼ각ᄒᆞ고졀을졀손으로ᄭᅮ지져왈:무난이졀졀의ᄎᆞᆫ소ᄅᆡᄉᆞᄅᆞᆷ이라ᄒᆞ고졔장vellous이라ᄒᆞ고ᄎᆞᄭᅮ지져보ᄂᆡ다ᄀᆞ지르니만일ᄉᆞᆷᄉᆞᆫ을졔ᄒᆞ면졀졀의다ᄀᆞᆺᄐᆞᆫ소ᄋᆞᆸ지아니ᄒᆞ리라ᄒᆞ고\n",
            "TOTAL TIME ELAPSED: 1.36s\n",
            "\n",
            "Gemma output:\n",
            "나는월궁의션녀로셔션븟노들고ᄂᆡ당의ᄒᆞᆫ경이되여젹상의ᄃᆡ셩이잇스되여젹군을ᄀᆞᆷ더니이졔졀의ᄒᆡᆼᄒᆞ여ᄃᆡ감의ᄀᆞᆫ감nis엇지잇스ᄒᆞ고ᄯᅩᄒᆞᆫᄃᆡ셩을장ᄒᆞ여ᄃᆡ감과ᄃᆡ감이잇스 lieve\n",
            "TOTAL TIME ELAPSED: 1.19s\n",
            "\n",
            "Gemma output:\n",
            "일쟝춘몽이라ᄃᆡᄀᆞᆷᄉᆞᆷ을이르지아니ᄒᆞ고ᄃᆡ감을감ᄉᆞ을ᄒᆞ야ᄃᆡ장셩을즁ᄒᆞ야ᄉᆞᆷᄉᆞᆷ을ᄯᅩᄒᆞᆫᄇᆡᆨ셩을ᄒᆞ야ᄉᆞᆷ도라ᄒᆞ니\n",
            "TOTAL TIME ELAPSED: 0.85s\n",
            "\n",
            "Gemma output:\n",
            "션녀왈그ᄃᆡᄉᆞᆫ은그ᄲᅮᆫ의ᄌᆡ믈이라\n",
            "TOTAL TIME ELAPSED: 0.27s\n",
            "\u001b[1m224/224\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m52s\u001b[0m 232ms/step - loss: 0.4975 - sparse_categorical_accuracy: 0.6059\n",
            "\n",
            "Gemma output:\n",
            "나랏말ᄊᆞ미건지라ᄒᆞ여분هور왈ᄀᆞ마음을ᄉᆡ esboço치고학ᄉᆡᆼ을모로며ᄭᅮ지져왈그 gracilis\n",
            "TOTAL TIME ELAPSED: 0.44s\n",
            "\n",
            "Gemma output:\n",
            "나는월궁의션녀로셔션즁의션러와즁슊을즁의잇셔으로즁의션즁의션러와즁슊을즁의잇셔셔으로즁의션즁의션러을\n",
            "TOTAL TIME ELAPSED: 0.65s\n",
            "\n",
            "Gemma output:\n",
            "일쟝춘몽이라ᄃᆡ지ᄂᆞᆫ바ᄀᆞ니ᄉᆞᆷ셩의다ᄀᆞᆷᄂᆡ여ᄃᆡ신졔ᄉᆞᄅᆞᆯᄒᆞ시고ᄃᆡ신ᄒᆞ여ᄃᆡ신ᄒᆞᆫ즁노ᄀᆞᆷᄒᆞ시니신ᄃᆡ풍우ᄀᆞᆫ부셰옵ᄐᆞᆫ지라\n",
            "TOTAL TIME ELAPSED: 0.94s\n",
            "\n",
            "Gemma output:\n",
            "션녀왈그ᄃᆡᄀᆞᆷ이ᄃᆡ지아니ᄒᆞ고ᄃᆡ감왈너희는너ᄂᆞᆫᄃᆡ을ᄀᆞ지지말고ᄃᆡ감이너희를ᄀᆞᆫᄃᆡᄒᆞ야쳔지쳬ᄀᆞᆫᄃᆡ을알ᄂᆞ며너희를ᄀᆞ여히ᄀᆞᄉᆞᆷ의평ᄉᆞ여도라ᄒᆞ니라\n",
            "TOTAL TIME ELAPSED: 1.08s\n",
            "\u001b[1m224/224\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m52s\u001b[0m 232ms/step - loss: 0.4890 - sparse_categorical_accuracy: 0.6118\n",
            "\n",
            "Gemma output:\n",
            "나랏말ᄊᆞ미건지ᄒᆞ여셕왕의ᄉᆞᆫ연ᄒᆞ여etcode치예ᄒᆞ면인ᄉᆞ의ᄉᆞᆫ을ᄌᆞ지 novioᄀᆞᆫ나라의오ᄉᆞ오ᄉᆞ오ᄉᆞ오ᄒᆞ고ᄉᆞᄅᆞᆷ을ᄌᆡ지 novio쳔ᄉᆡᆼᄒᆞ여오더니오ᄉᆞ오ᄉᆞ오더ᄒᆞᆫ지라\n",
            "TOTAL TIME ELAPSED: 0.97s\n",
            "\n",
            "Gemma output:\n",
            "나는월궁의션녀로셔션즁의션즁을ᄃᆡᄃᆡᄭᅳᆯ셔dtos을ᄯᅳ셔로ᄯᅳᆫ셔ᄌᆡ즈을드러ᄀᆞ리업더니ᄃᆡ정ᄒᆞᆫ이ᄀᆞ로ᄃᆡ풍우ᄀᆞᆺᄐᆞᆫ지라\n",
            "TOTAL TIME ELAPSED: 0.82s\n",
            "\n",
            "Gemma output:\n",
            "일쟝춘몽이라ᄒᆞᆫ바니니ᄒᆞᆫᄉᆞᄅᆞᆷ이ᄌᆡ조길동을ᄌᆞᄇᆞ드리라ᄒᆞ고ᄌᆡ조이번왕조슈ᄅᆞᆯᄉᆡᆼᄂᆡ라ᄒᆞ니\n",
            "TOTAL TIME ELAPSED: 0.63s\n",
            "\n",
            "Gemma output:\n",
            "션녀왈그ᄃᆡᄀᆞᆷ이아니요ᄒᆞᄂᆞᆫ거ᄂᆞᆫ슈로다시ᄯᅩᄒᆞᆫᄌᆡ홈을ᄯᅩᄒᆞᆫᄌᆡ슈로ᄒᆞ여ᄌᆡ어귀의아온을일시의ᄅᆞᄀᆞᄒᆞᄂᆞ이다ᄒᆞ고어릇어릇ᄒᆞ여ᄌᆡ젹을ᄯᅩᄒᆞᆫᄌᆡ슈로ᄒᆞ여어미을ᄯᅩᄒᆞ니라\n",
            "TOTAL TIME ELAPSED: 1.25s\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "# Inference after fine tuning"
      ],
      "metadata": {
        "id": "sHURthL7HCiR"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "text_gen(\"나랏말ᄊᆞ미\")\n",
        "text_gen(\"나는월궁의션녀로셔\")\n",
        "text_gen(\"일쟝춘몽이라\")\n",
        "text_gen(\"션녀왈\")"
      ],
      "metadata": {
        "id": "DddIrGsvHF-r",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "1538352a-e129-4f97-f3b0-a0add94e4bd8"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "Gemma output:\n",
            "나랏말ᄊᆞ미 ᄒᆞᄉᆞ 율왕의 ᄒᆞᆫ ᄉᆞᄅᆞᆷ의 몸을 ᄉᆞᄅᆞᆷ 막고 왕실의 ᄒᆞᆫ임ᄀᆞ ᄒᆞᆫ지라. 율왕의 몸을 ᄉᆞᄅᆞᆷ이 잇셔 ᄒᆞᆫ ᄉᆞᄅᆞᆷ이 잇셔 ᄒᆞᆫ지라. ᄒᆞᆫ이라 ᄒᆞᆫ 율왕의 몸을 ᄉᆞᄅᆞᆷ의 ᄒᆞᆫ ᄉᆞᄅᆞᆷ이 잇셔 ᄒᆞᆫ지라. 율왕의 몸을 ᄉᆞᄅᆞᆷ이 잇셔 ᄒᆞᆫ지라. 율왕의 몸을 ᄅᆞᄅᆞᆷ이 잇셔 ᄒᆞᆫ지라. 율왕의 몸을 ᄉᆞᄅᆞᆷ이 잇셔 ᄒᆞᆫᄉᆡᆨ ᄯᆞ히ᄒᆞᆫ 몸이 ᄂᆡ난지라. 율왕의 몸을 ᄃᆡᄉᆞᆷ이 잇셔 ᄒᆞ\n",
            "TOTAL TIME ELAPSED: 10.52s\n",
            "\n",
            "Gemma output:\n",
            "나는 월궁의 션녀로셔 ᄌᆞᆸ기 ᄌᆞᄌᆞᆷ의 길을 ᄉᆞᆷEtter이 ᄒᆞᆫ셔 ᄒᆞᆫ셔 ᄌᆞᄌᆞᆷ의 길을 ᄌᆞᄌᆞᆷ의 ᄌᆞᄇᆞ셔셔 ᄒᆞᆫ셔 ᄇᆡᆨ셩을 ᄉᆞᄅᆞᆷ의 길을 ᄌᆞᄇᆞ셔셔 ᄒᆞᆫ셔 ᄇᆡᆨ셩을 ᄉᆞᄅᆞᆷ으로 ᄒᆞᆫ셔 ᄇᆡᆨ셩의 ᄇᆡ을 ᄉᆞᄅᆞᆷ으로 ᄌᆞᄇᆞ셔셔 ᄇᆡᆨ셩의 길을 ᄒᆞᆫ셔 ᄇᆡ을 ᄯᅥᄒᆞᆫ셔 ᄇᆡᆨ셩의 ᄇᆡ을 ᄉᆞᄅᆞᆷ으로 ᄌᆞᄌᆞᆷ의 길을 ᄇᆞᄅᆞ셔셔 ᄇᆡᆨ셩의 길을 ᄇᆡ기 ᄌᆞᄇᆞ셔셔 ᄇᆡᆨ�\n",
            "TOTAL TIME ELAPSED: 2.54s\n",
            "\n",
            "Gemma output:\n",
            "일쟝춘몽이라 ᄒᆞᆫ쳔 ᄌᆡ상이 ᄒᆞᆫ고 ᄉᆞᄅᆞᆷ의 ᄒᆞᆫ고 ᄒᆡᆼᄒᆞᆯ 잇셔 닷ᄀᆞᆺ치 ᄒᆞᆫ셔 ᄉᆞᄅᆞᆫ 군ᄉᆞ ᄀᆞᆺ치 ᄒᆞᆫ셔 ᄉᆞᄅᆞᆷ의 ᄌᆡ조 ᄒᆞᆫ군이 ᄒᆞᆫ고 ᄉᆞᄅᆞᆷ의 ᄒᆞᆫ고 ᄒᆞᆫᄉᆡᆼ ᄉᆞᄅᆞᆷ이 이르러 ᄀᆞᆺ쳐 ᄒᆞᆫ셔 ᄉᆞᄅᆞᆷ의 집히 ᄒᆞᆫ셔 ᄉᆞᄅᆞᆷ은 ᄒᆞᆫ셔 ᄀᆞᆡ의 ᄯᆞᄅᆞᆫ이 이러ᄒᆞᆫ ᄀᆞᆺ치 ᄒᆞᆫ듯이 ᄒᆞᆫ듯 ᄯᆡ ᄯᅬᄒᆞᆫ ᄉᆞᄅᆞᆷ이 ᄒᆞᆫ셔 ᄯᅩᄒᆞᆫ ᄉᆞᄅᆞᆷ이 이러ᄒᆞ\n",
            "TOTAL TIME ELAPSED: 2.57s\n",
            "\n",
            "Gemma output:\n",
            "션녀왈 ᄉᆞᆷ졔 길의 ᄌᆞᆷ수을 ᄉᆞᆷ고, ᄃᆡ장부 ᄌᆞᆷ수을 ᄌᆞᆷ수으로 바꾸 ᄯᆞᆫ지 말고 ᄒᆞᆫ ᄉᆞᆷ수을 ᄌᆞᆷ수으로 바꾸 ᄯᆞᆫ지라. ᄃᆡ장부 ᄒᆞᆫ ᄉᆞᆷ수을 ᄉᆞᆷ수으로 바꾸 ᄯᆞᆫ지 말고 ᄇᆡᆨ셩을 ᄌᆞᆷ수을 ᄒᆞᆫ지 말고 ᄇᆡᆨ셩을 ᄒᆞᆫ지 말 ᄒᆞ야 ᄒᆞᆫ ᄉᆞᆷ을 ᄌᆞᆷ수을 ᄃᆡᄒᆞ야 ᄃᆡ장부 ᄇᆞᆫ ᄉᆞᆷ을 ᄉᆞᆷ수으로 바꾸 �\n",
            "TOTAL TIME ELAPSED: 2.12s\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "# Load pre-trained LoRA"
      ],
      "metadata": {
        "id": "xdhsydAWGv_C"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "gemma.backbone.enable_lora(lora_rank)\n",
        "gemma.backbone.load_lora_weights(f\"/content/drive/MyDrive/{lora_name}_{lora_rank}.lora.h5\")"
      ],
      "metadata": {
        "id": "aAYpaUY6GvD5"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "gemma.compile(sampler=keras_nlp.samplers.TopKSampler(k=3, temperature=0.7))"
      ],
      "metadata": {
        "id": "9Y8bSnHVHtBD"
      },
      "execution_count": null,
      "outputs": []
    }
  ]
}