LLM_CA4_RLHF.ipynb

{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "a9AWeWaSliGs"
      },
      "source": [
        "## CA 4, LLMs Spring 2024\n",
        "\n",
        "- **Name:** Mohammad Mahdi Salmani\n",
        "- **Student ID:** 810102174\n",
        "\n",
        "---\n",
        "### This is due on **14 June**, submitted via [elearn](https://elearn.ut.ac.ir/).\n",
        "#### Your submission should be named using the following format: `CA4_LASTNAME_STUDENTID.ipynb`.\n",
        "\n",
        "---\n",
        "\n",
        "##### *How to do this problem set:*\n",
        "\n",
        "- Some questions require writing Python code and computing results, and the rest of them have written answers. For coding problems, you will have to fill out all code blocks that say `WRITE YOUR CODE HERE`.\n",
        "\n",
        "- For text-based answers, you should replace the text that says \"Write your answer here...\" with your actual answer.\n",
        "\n",
        "- There is no penalty for using AI assistance on this homework as long as you fully disclose it in the final cell of this notebook (this includes storing any prompts that you feed to large language models). That said, anyone caught using AI assistance without proper disclosure will receive a zero on the assignment (we have several automatic tools to detect such cases). We're literally allowing you to use it with no limitations, so there is no reason to lie!\n",
        "\n",
        "---\n",
        "\n",
        "##### *Academic honesty*\n",
        "\n",
        "- We will audit the Colab notebooks from a set number of students, chosen at random. The audits will check that the code you wrote actually generates the answers in your notebook. If you turn in correct answers on your notebook without code that actually generates those answers, we will consider this a serious case of cheating.\n",
        "\n",
        "- We will also run automatic checks of Colab notebooks for plagiarism. Copying code from others is also considered a serious case of cheating.\n",
        "\n",
        "---"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "TqjNIhG0liGv"
      },
      "source": [
        "# RLHF (55 points)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "qTo0AfIPliGv"
      },
      "source": [
        "## Introduction to RLHF"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "hOQLOhnMliGv"
      },
      "source": [
        "<img src=\"https://d2908q01vomqb2.cloudfront.net/f1f836cb4ea6efb2a0b1b99f41ad8b103eff4b59/2023/08/31/ML-14874_image001.jpg\"/>\n",
        "</div>\n",
        "\n",
        "With the recent public introduction of ChatGPT, reinforcement learning from human feedback (RLHF) has become a hot topic in language modeling circles -- both academic and industrial.\n",
        "We can trace the application of RLHF to natural language processing OpenAI's 2019 release of <br>[Fine-Tuning Language Models from Human Preferences](https://arxiv.org/abs/1909.08593).\n",
        "\n",
        "Fast forward one year when OpenAI released one of its first significant papers on reinforcement learning from human feedback applied to natural language generation.\n",
        "\n",
        "In that paper-<br>[Learning to summarize from human feedback](https://arxiv.org/abs/2009.01325)-OpenAI showed that simply fine-tuning on summarization data leads to suboptimal performance when evaluated on human preferences. The authors suggest optimizing for human preferences directly via a reinforcement learning approach to alleviate these performance issues.\n",
        "\n",
        "\n",
        "**Learn More:**\n",
        "<br>[Huggingface Deep Reinforcement Learning Course](https://huggingface.co/learn/deep-rl-course/en/unit0/introduction)\n",
        "<br>[Research Papers for Reinforcement Learning with Human Feedback ](https://github.com/opendilab/awesome-RLHF)\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "qWJejByfliGv"
      },
      "source": [
        "## Import Libraries and Set Constants"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "zIqcvJY4liGv"
      },
      "outputs": [],
      "source": [
        "%pip install datasets\n",
        "%pip install evaluate\n",
        "%pip install rouge_score\n",
        "%pip install accelerate -U\n",
        "%pip install transformers[torch]"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 1,
      "metadata": {
        "id": "K0uv-c3HliGw"
      },
      "outputs": [],
      "source": [
        "import numpy as np\n",
        "import pandas as pd\n",
        "import json\n",
        "import random\n",
        "import evaluate\n",
        "import matplotlib.pyplot as plt\n",
        "import os\n",
        "import torch\n",
        "import torch.nn as nn\n",
        "import torch.optim as optim\n",
        "from datasets import load_dataset\n",
        "from torch.utils.data import Dataset, DataLoader\n",
        "import transformers\n",
        "from transformers import AutoModelForSequenceClassification, AutoTokenizer, get_scheduler\n",
        "from transformers import (\n",
        "    AutoModelForCausalLM,\n",
        "    AutoTokenizer,\n",
        "    Trainer,\n",
        "    TrainingArguments,\n",
        "    default_data_collator,\n",
        ")\n",
        "from transformers import AdamW\n",
        "from tqdm import tqdm\n",
        "import warnings\n",
        "warnings.filterwarnings(\"ignore\")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 2,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 177,
          "referenced_widgets": [
            "e400bbe893d744b4bc4ec3c00d0ab49a",
            "57be34f94b554f0e94f45635efd721b7",
            "04588e43dc174ae5bcebc0ef01d581dc",
            "bd1e9d756633421e980ac84a1ed3df96",
            "ed839cf80d8e4f4a8feaa109cd9202b6",
            "7cf3fe343cdd45b291523b15dc8f4c1a",
            "ac236fc0d5bc4d8f962e3d35b859893d",
            "58cabc7c155747fc8cbe2f27f1beb289",
            "cce757f48382449da0d3e3bf87272284",
            "99d0e034709c40359095a9004cc46798",
            "e6953b8a8ace427ca7689f7fc6cfd6b3",
            "a22888ab49b941bfae153b1d7d6a3fe9",
            "054c5f4fc64744a7a46e991855c7b386",
            "caaca6163cff4afb9b28978b1e64065a",
            "896bed8372804158aa4bc944e6401f42",
            "40465bc3852f40018eb5173d2a789c7f",
            "d18ab3331f1a42f1af89aca50b9e78e6",
            "08a9f29c73b54e0cbc89738191edb1a8",
            "12ce1164d18249b3807e5ec45aac4a84",
            "1f99bdeb99dc43f882494cb9a8d1f664",
            "41f16b59ab674b67ab1577adb413ae48",
            "1dd9d180974942d89cabcc4d318a6cf0",
            "b41bb97a0a7a4cd589d837144eed8200",
            "5446c6060f8e419ab8184d0e15981690",
            "236c712f6f234e3f8d7f4ea00feb7f6a",
            "7b770a26b25c449fbdac9824c6b00204",
            "30ae280e5d5b462daa1f61b929e51a3e",
            "a97ec2a948e549ae9d97c2ce6d8d962e",
            "b2729e2fde2549d1b531e9320dd6649d",
            "e8bd37b87fba4f8a9e2d0e9bb80a13f0",
            "748e4fce751841348c8907d56a01d2a1",
            "8cc35a3254b049de9d1535e4c4a64dda",
            "5835fb83bd014edcaedc1974e2738273",
            "7cd8e757c96f4a4481c5fffd99fb2cf2",
            "c5541f7a0b3d405388df008eb820a130",
            "ca480988b2404679954759cb9e2c991c",
            "f73da3dc44384c7baad205eaa14bfb41",
            "af155570f61742c181961e49703cf911",
            "9b3cc7be38ab486abb29f68c4ec58388",
            "5e9e490c3c244b09a1d8f9007a5d8717",
            "97816036a7c643feb7c24755f4e2233b",
            "8ee845415acf4bb29a6a3fc5c2c72070",
            "41afb58c5a964028a4943def4b861b97",
            "55634652e74049b199344e325dff5676",
            "dfc95e5b17b94e1dae5b0c5a26a9da5d",
            "7bcdb14ff8c64ff4b1d7b882a85f89dd",
            "2e3fa17580e544aaa238a07cdeb116b7",
            "723a0a0fae6344b2bdc445d69af1517a",
            "29c0f9d47313413386f284f7393211d1",
            "005b8f782bdb4ab2a8a5e40a75674d96",
            "5c38c56a72d74118a65c7eca2d47aaa0",
            "af5827003e7c43d8b6bda1efceeb1fae",
            "5d82735ec2df45c7b9fd820df07d938d",
            "c16c6f1424134cb7ab354757b7ce2c15",
            "0b831cc698eb4838960af0add2becea1"
          ]
        },
        "id": "dwFPQ6UFliGw",
        "outputId": "105e42f6-831e-42a5-e6f9-39d11c7daf9c"
      },
      "outputs": [],
      "source": [
        "class CONFIG:\n",
        "    seed = 42\n",
        "    max_len = 550\n",
        "    train_batch_size = 16\n",
        "    eval_batch_size = 1\n",
        "    eval_steps = 500\n",
        "    epochs = 5\n",
        "    save_steps = 1000\n",
        "    learning_rate = 1e-4\n",
        "    gradient_accumulation_steps = 1\n",
        "    model_name = 'gpt2'\n",
        "    tokenizer = AutoTokenizer.from_pretrained(model_name)\n",
        "    device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')\n",
        "    output_dir = \"gpt2-supervised-summarize-checkpoint\"\n",
        "    output_dir_rm = \"rm_checkpoint\"\n",
        "\n",
        "device = CONFIG.device\n",
        "rw_device = CONFIG.device"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "snknIeG6liGw"
      },
      "source": [
        "## Implementing Learning for Summarization"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "dBOWhTBJliGx"
      },
      "source": [
        " In this notebook by using trlX, we will implement RLHF for a summarization task. The training process consists of three parts:\n",
        "\n",
        "*   We will first fine-tune a pre-trained transformer model on our summarization dataset. This is our supervised fine-tuned model (SFT).\n",
        "* We will then train a reward model (RM). This model is initialized from the SFT model and outputs a scalar value. This scalar value is the reward that indicates the preferability of a summary.  \n",
        "\n",
        "*   Finally, we use the RM to fine-tune the SFT model via PPO. This step aligns our SFT model with human preference."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "nXsS0ZcgliGx"
      },
      "source": [
        "## Section One: Supervised Fine Tuning (5 points)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "66TUpjRyliGx"
      },
      "source": [
        "### Dataset"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "hJ1IjMDQliGx"
      },
      "source": [
        "For our experiment, we'll use the **TLDR summarization** dataset used originally in Learning to summarize from human feedback.\n",
        "\n",
        "Based on that training process described above, we'll need two types of datasets:\n",
        "\n",
        "*   One for fine-tuning the pre-trained supervised model and then for fine-tuning it again with PPO and reward model, and\n",
        "*   One for training our reward model.\n",
        "\n",
        "In our case, the dataset for fine-tuning is the filtered* TLDR dataset. The dataset for training our reward model is the **comparison or preference dataset**.\n",
        "\n",
        "**Note:** I set the number of training examples to 6000, you can increase it, also you can adjust the number of validation examples.\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 3,
      "metadata": {
        "id": "yJJ6gvoKliGx"
      },
      "outputs": [],
      "source": [
        "tlrdataset_path = \"CarperAI/openai_summarize_tldr\"\n",
        "comparissions_path = \"CarperAI/openai_summarize_comparisons\""
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "6yI8w7wuliGx"
      },
      "source": [
        "#### Create Dataset"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 4,
      "metadata": {
        "id": "lJW3S1KpliGx"
      },
      "outputs": [],
      "source": [
        "class TLDRDataset(Dataset):\n",
        "    def __init__(self, path, tokenizer, split, max_length=CONFIG.max_len):\n",
        "        self.post_list = []\n",
        "        dataset = load_dataset(path, split=split)\n",
        "        for sample in dataset:\n",
        "            self.post_list.append(sample[\"prompt\"] + sample[\"label\"])\n",
        "\n",
        "        if \"train\" in split:\n",
        "          self.post_list = random.sample(self.post_list, min(6000, len(self.post_list)))\n",
        "        elif \"valid\" in split:\n",
        "            self.post_list = self.post_list[0:2000]\n",
        "\n",
        "        self.tokenizer = tokenizer\n",
        "        self.max_length = max_length\n",
        "        self.input_ids = []\n",
        "        self.attn_masks = []\n",
        "\n",
        "    def __len__(self):\n",
        "        return len(self.post_list)\n",
        "\n",
        "    def __getitem__(self, idx):\n",
        "        txt = self.post_list[idx]\n",
        "        encodings_dict = self.tokenizer(txt, truncation=True, max_length=self.max_length, padding=\"max_length\")\n",
        "        input_ids = torch.tensor(encodings_dict[\"input_ids\"])\n",
        "        attn_masks = torch.tensor(encodings_dict[\"attention_mask\"])\n",
        "\n",
        "        return {\n",
        "            \"input_ids\": input_ids,\n",
        "            \"attention_mask\": attn_masks,\n",
        "            \"labels\": input_ids,\n",
        "        }"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "sjxXIrRyliGx"
      },
      "source": [
        "#### Load Dataset"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 5,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 241,
          "referenced_widgets": [
            "caa2f6b29a0349aa80d773185181056b",
            "8d354fed57014964962451731bb319ef",
            "0faf1d3472974ceca6eeda7e9d5f7906",
            "b54c45a49eb1484194a576d0421c0fe0",
            "2bb6760048f24cb59f6ff5f16b3c7acd",
            "93c404b763094295aa1bbcdc5847e3b9",
            "2c3ffd876324474594f95fe1c71d3782",
            "c6c5a944a23e456ebb6c69805e4d4e2e",
            "196db38e52bf47e0a529d725ad71d007",
            "a483b3f8216d4de88866cb1a017c14bb",
            "3451ffe86a4a4d7a8858e4e720789857",
            "5d1aaf2a41fd47e1ba3a2a73855db66b",
            "4a9bad66b87348589618e074a6b2f747",
            "94712aabc9504ec7a3d204a914118b6c",
            "70010175a0c9460c89a2297895f9425e",
            "9cf51736a40e490ba9daa0aa4585d614",
            "5cef8de414a447ff911632ff05beedfb",
            "39e1e4fa159044ea97c5f9c78ba14a1c",
            "45a46219b04b4d5989b5a7a241c165b5",
            "a5c0c29181fc4821b4017e7028372431",
            "f61f3b244ab5421e8a2c18d37c33e290",
            "920ac1c6300d4ac2b826fa0fcc1c4cd8",
            "458e2312cf2c47fea736fc230083a0de",
            "83f9dbdfcf7443ce8589e0c59116afcf",
            "9ed93d180d26497596cdeb3f8d014e09",
            "9544135d58ee491191ea3b16bf2424eb",
            "2510e83f724343019d01a35a84e76bf9",
            "c8cc0f0cf5d14e83b002ead2a0c3afa4",
            "de3fe2d2e22547e1beb6c76b59d71082",
            "d1fe338ae4604838b403b255a4637154",
            "5243aeab7e624b02b273463fb32fbf42",
            "90066549c0b4446bbae7b2a16fbe1a6b",
            "18439c99c16d4162b8a2a52d86dc5847",
            "b6fa7a2add97484cbea0c26ed46d29e6",
            "44e178a0be524daa95b256d5bff31202",
            "c33b7e89df304f78b3b498a8f9c68383",
            "ac756657854e4ec78993df53314cc18f",
            "2923856e65fd40718b7358bfbb3143cf",
            "452cea6360ce498c85e843e348deb1f4",
            "16f10c9d3be140829e228ad4b3aa7705",
            "bda44f920ef0452991f345ffb8843f73",
            "890cde66805f46b2bcc713df00dfbe22",
            "ed16f76131904795900803fb1238f73d",
            "334a2eeec579455ba662deabd578f29d",
            "d3fccbdf94bc44cdb552e3ac37a54e0f",
            "ddcd794463174f51b09f6b77ab732b94",
            "4759413c86e845d9a8e374ee135ff414",
            "38b91d003fc541908101453b9655abb2",
            "7d3ece0ec61a4f7eb21808f759df7690",
            "29201ca978c5470aa05877d0ace36f31",
            "7151d446f6964946af694275001a8b50",
            "ab0a3df7a8904389904bfcb31a148b7f",
            "c692217a3ef84e04bccb3bcdd4a2ec8a",
            "d6ba35709a0f4221bff3209b66f6ccde",
            "64f0e8ad62e448f9b767515e44bf4e9b",
            "329fab4cb7a14053b47bdd56ad38ec60",
            "d839e15480684df99f1ce3a7c6c41342",
            "774a79f0517a4d5c8b2a037f6a59dc3e",
            "7b52f480da1e4554a39b731269dc5d28",
            "96762776b2564b0ca0eda807d998d097",
            "d7554a7bb8bd4d5dbbc03268dc2d77c4",
            "cf4a4ec0a7124457a68d70970889b744",
            "4929031e60fa4ad883215fc9470113ae",
            "a6bc156f830b4829b35f298ed7d733bf",
            "6e2efc4590694d2ab3e02f4baa2ed821",
            "de83309c1bd54e66a613b755a89f4455",
            "5d88dfa6c6e04cc689bdc9f753fd26fa",
            "bb4c6b6b74824b6b8fcc74009aa36768",
            "c6acfb70629745729fa440a02b5aa6f1",
            "d4c3590ed00243a99495add140c2f504",
            "4b02a2cb6fe14ccd9d85a86994c6cb67",
            "a4d54d72871743a68f60983a3bdbf55d",
            "0e017c1b69fe4e89b3a5e169c9387445",
            "5dd2682ec2ab4a1ba0d0023ffe3cc4a9",
            "4a452ec981ff415680f80c2cfd30a0ad",
            "e53325f28cab44a5ab6a9218764b45d2",
            "d5f193695fde47e8b9edd4f84f2ecf2b"
          ]
        },
        "id": "KcTIsUmfliGx",
        "outputId": "6b2d1543-b17d-442c-fabf-37b228ff4e23"
      },
      "outputs": [],
      "source": [
        "train_dataset = TLDRDataset(\n",
        "  path=tlrdataset_path,\n",
        "  tokenizer=CONFIG.tokenizer,\n",
        "  split=\"train\")\n",
        "\n",
        "dev_dataset = TLDRDataset(\n",
        "  path=tlrdataset_path,\n",
        "  tokenizer=CONFIG.tokenizer,\n",
        "  split=\"valid\")"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "kGyPTscSliGx"
      },
      "source": [
        "### Load Model and Tokenizer"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 20,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 81,
          "referenced_widgets": [
            "ff6e07c2c9de408cb353b2b57ad52673",
            "8525e5dc75cb42d790ee5e59c96cee5f",
            "f10cd68089cc4e3ebe9b774054517c80",
            "27fa70cf544147f29b2a8a17c4f2faa3",
            "fe88b66639f64622adda4094cb268fa3",
            "19a4391701ad4399887e35757ed7eea6",
            "7b6d4a6e48be478a8f694bccbdfaa81d",
            "646ee18fffee4f0c89f23871c56223a8",
            "8cb3dd49227d4a72b4d7938444f0202a",
            "64f492d1096149ad9da667ea864d7f2b",
            "bc9c54b508ed4aabad64135937d16a70",
            "2053f4c66e57485a8e6176c4dbb68bd1",
            "7691a29d5bf44d01bafd6643400cbef0",
            "ba580709444348ac99daf82d488239a3",
            "5ddf722463404dbca2e1343722d2fe6b",
            "d4491208f1d641adbba7e3d9138aab5a",
            "8bbdbacf56ee44aca93e9a9a7ea824e0",
            "711cce8380d8411181387e51b7da25bb",
            "bf8df483688b4650bb5279a23602a520",
            "dbe917ae193a4c3c8cbcc9254dd8f639",
            "e47ec19a85da4dbaab999b5bb6490c05",
            "2e3a568ae6714c4aab97b67c7db2eb07"
          ]
        },
        "id": "2Hsmbn4QliGx",
        "outputId": "c3e97638-4673-4b67-f802-812d5c959544"
      },
      "outputs": [],
      "source": [
        "#Load the model and tokenizer\n",
        "model = AutoModelForCausalLM.from_pretrained(CONFIG.model_name, use_cache=False)\n",
        "tokenizer = CONFIG.tokenizer\n",
        "\n",
        "# Setting pad token to eos token\n",
        "tokenizer.pad_token = tokenizer.eos_token\n",
        "tokenizer.pad_token_id = tokenizer.eos_token_id\n",
        "\n",
        "# Resize token embeddings\n",
        "model.resize_token_embeddings(len(tokenizer))\n",
        "\n",
        "# Update model configuration\n",
        "model.config.pad_token_id = tokenizer.pad_token_id\n",
        "model.config.end_token_id = tokenizer.eos_token_id"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "S_BfUJkMliGx"
      },
      "source": [
        "### Define Compute metric function (2.5 Points)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "mNRBB004liGx"
      },
      "source": [
        "In this part, you should implement an evaluation function that computes rouge scores for our predicted summaries."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 16,
      "metadata": {
        "id": "oAUXiZIzliGx"
      },
      "outputs": [],
      "source": [
        "# Load the ROUGE metric\n",
        "rouge = evaluate.load(\"rouge\")\n",
        "\n",
        "def compute_metrics(eval_preds):\n",
        "    preds, labels = eval_preds\n",
        "    decoded_preds = [tokenizer.decode(pred, skip_special_tokens=True, clean_up_tokenization_spaces=True) for pred in preds]\n",
        "    decoded_labels = [tokenizer.decode(label, skip_special_tokens=True, clean_up_tokenization_spaces=True) for label in labels]\n",
        "    result = rouge.compute(predictions=decoded_preds, references=decoded_labels)\n",
        "    return result"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 17,
      "metadata": {
        "id": "4PN_qjWOliGy"
      },
      "outputs": [],
      "source": [
        "# Create a preprocessing function to extract out the proper logits from the model output\n",
        "def preprocess_logits_for_metrics(logits, labels):\n",
        "    if isinstance(logits, tuple):\n",
        "        logits = logits[0]\n",
        "    return logits.argmax(dim=-1)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "W6lnAyR4liGy"
      },
      "source": [
        "### Train"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 21,
      "metadata": {
        "id": "TPpjeWjqliGy"
      },
      "outputs": [],
      "source": [
        "# Set up training arguments\n",
        "training_args = TrainingArguments(\n",
        "    output_dir=CONFIG.output_dir,\n",
        "    evaluation_strategy=\"steps\",\n",
        "    eval_steps=500,\n",
        "    save_steps=1000,\n",
        "    eval_accumulation_steps=1,\n",
        "    learning_rate=CONFIG.learning_rate,\n",
        "    per_device_train_batch_size=CONFIG.train_batch_size,\n",
        "    per_device_eval_batch_size=CONFIG.eval_batch_size,\n",
        "    gradient_checkpointing=True,\n",
        "    fp16=True,\n",
        "    fp16_backend=\"auto\",\n",
        "    adam_beta1=0.9,\n",
        "    adam_beta2=0.95,\n",
        "    num_train_epochs=1,\n",
        "    warmup_steps=100,\n",
        "    load_best_model_at_end=True,\n",
        "    logging_steps=50\n",
        ")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 22,
      "metadata": {
        "id": "q4zCE76CliGy"
      },
      "outputs": [],
      "source": [
        "# Initialize the trainer\n",
        "trainer = Trainer(\n",
        "    model=model,\n",
        "    args=training_args,\n",
        "    train_dataset=train_dataset,\n",
        "    eval_dataset=dev_dataset,\n",
        "    compute_metrics=compute_metrics,\n",
        "    data_collator=default_data_collator,\n",
        "    preprocess_logits_for_metrics=preprocess_logits_for_metrics\n",
        ")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 23,
      "metadata": {},
      "outputs": [],
      "source": [
        "torch.cuda.empty_cache()"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 36,
      "metadata": {},
      "outputs": [
        {
          "data": {
            "text/html": [
              "\n",
              "    <div>\n",
              "      \n",
              "      <progress value='752' max='752' style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
              "      [752/752 37:12, Epoch 4/4]\n",
              "    </div>\n",
              "    <table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              " <tr style=\"text-align: left;\">\n",
              "      <th>Step</th>\n",
              "      <th>Training Loss</th>\n",
              "      <th>Validation Loss</th>\n",
              "      <th>Rouge1</th>\n",
              "      <th>Rouge2</th>\n",
              "      <th>Rougel</th>\n",
              "      <th>Rougelsum</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <td>250</td>\n",
              "      <td>1.898400</td>\n",
              "      <td>1.923292</td>\n",
              "      <td>0.582104</td>\n",
              "      <td>0.180484</td>\n",
              "      <td>0.381413</td>\n",
              "      <td>0.504274</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>500</td>\n",
              "      <td>1.875600</td>\n",
              "      <td>1.912207</td>\n",
              "      <td>0.586530</td>\n",
              "      <td>0.182838</td>\n",
              "      <td>0.383275</td>\n",
              "      <td>0.509494</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>750</td>\n",
              "      <td>1.843300</td>\n",
              "      <td>1.908619</td>\n",
              "      <td>0.586853</td>\n",
              "      <td>0.183296</td>\n",
              "      <td>0.384129</td>\n",
              "      <td>0.509715</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table><p>"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        },
        {
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "There were missing keys in the checkpoint model loaded: ['lm_head.weight'].\n"
          ]
        }
      ],
      "source": [
        "# Start training\n",
        "trainer.train()\n",
        "\n",
        "# Save the model\n",
        "trainer.save_model(CONFIG.output_dir)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 24,
      "metadata": {},
      "outputs": [
        {
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "\u001b[34m\u001b[1mwandb\u001b[0m: Logging into wandb.ai. (Learn how to deploy a W&B server locally: https://wandb.me/wandb-server)\n",
            "\u001b[34m\u001b[1mwandb\u001b[0m: You can find your API key in your browser here: https://wandb.ai/authorize\n",
            "\u001b[34m\u001b[1mwandb\u001b[0m: Paste an API key from your profile and hit enter, or press ctrl+c to quit:"
          ]
        }
      ],
      "source": [
        "trainer.train()\n",
        "trainer.save_model(CONFIG.output_dir)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 37,
      "metadata": {},
      "outputs": [
        {
          "data": {
            "text/html": [
              "\n",
              "    <div>\n",
              "      \n",
              "      <progress value='268' max='63' style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
              "      [63/63 04:36]\n",
              "    </div>\n",
              "    "
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        },
        {
          "data": {
            "text/plain": [
              "{'eval_loss': 1.9122065305709839,\n",
              " 'eval_rouge1': 0.5865301790188278,\n",
              " 'eval_rouge2': 0.1828376831114582,\n",
              " 'eval_rougeL': 0.383274792047039,\n",
              " 'eval_rougeLsum': 0.509493881204031,\n",
              " 'eval_runtime': 128.17,\n",
              " 'eval_samples_per_second': 15.604,\n",
              " 'eval_steps_per_second': 0.492,\n",
              " 'epoch': 4.0}"
            ]
          },
          "execution_count": 37,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "trainer.evaluate()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "7UhxFCAwliGy"
      },
      "source": [
        "### Test (2.5 Points)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "rVHkMLrMliGy"
      },
      "source": [
        "Report rouge scores for test set of TLDR dataset."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 38,
      "metadata": {},
      "outputs": [],
      "source": [
        "test_dataset = TLDRDataset(\n",
        "    path=tlrdataset_path,\n",
        "    tokenizer=CONFIG.tokenizer,\n",
        "    split=\"test\"\n",
        ")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 39,
      "metadata": {},
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Test ROUGE scores:\n"
          ]
        },
        {
          "data": {
            "text/plain": [
              "{'eval_loss': 1.9079687595367432,\n",
              " 'eval_rouge1': 0.5875083782980206,\n",
              " 'eval_rouge2': 0.1831651352260253,\n",
              " 'eval_rougeL': 0.3838607240070928,\n",
              " 'eval_rougeLsum': 0.509185929201362,\n",
              " 'eval_runtime': 426.4569,\n",
              " 'eval_samples_per_second': 15.366,\n",
              " 'eval_steps_per_second': 0.481,\n",
              " 'epoch': 4.0}"
            ]
          },
          "execution_count": 39,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "test_results = trainer.evaluate(test_dataset)\n",
        "print(\"Test ROUGE scores:\")\n",
        "test_results"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "bLQOsyhCliGy"
      },
      "source": [
        "## Section Two: Reward Model Training (25 Points)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "kWwhqfQdliGy"
      },
      "source": [
        "### Reward Model"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "DxD2B787liGy"
      },
      "source": [
        "We'll initialize the reward model from the SFT model and attach a randomly initialized linear head that outputs a scalar value on top.\n",
        "\n",
        "Next, we'll dig into how the data is input to the model, the loss function, and other gotchas of a reward model in more detail."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "PRdMjFNlliGy"
      },
      "source": [
        "### Question 1 (2.5 Points)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "PzaalFlCliGy"
      },
      "source": [
        "**How would you create a comparison dataset for a text summarization task? (explain the entire procedure)**"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "PgnAERZHliGy"
      },
      "source": [
        "This thask involves several key steps: data collection, preprocessing, initial summarization, human annotation, and dataset structuring. Here’s a detailed guide:\n",
        "\n",
        "**1. Data Collection**\n",
        "- Source Selection: Choose diverse sources like news articles, research papers, or blogs.\n",
        "- Data Gathering: Collect a substantial number of documents through web scraping, APIs, or open datasets.\n",
        "\n",
        "**2. Preprocessing**\n",
        "- Text Cleaning: Remove unwanted characters and normalize the text.\n",
        "- Segmentation: Break documents into smaller units if they are too long.\n",
        "\n",
        "**3. Initial Summarization**\n",
        "- Automatic Summarization: Use the model to generate initial summaries for each document.\n",
        "\n",
        "**4. Human Annotation**\n",
        "- Annotation Guidelines: Develop clear guidelines for comparing summaries on relevance, coherence, conciseness, readability, and accuracy.\n",
        "- Annotator Training: Train annotators to apply guidelines consistently.\n",
        "- Pairwise Comparison: Present annotators with pairs of summaries (e.g., machine vs. human-written) and ask them to rate or rank them.\n",
        "- Rating and Ranking: Collect ratings or rankings based on the provided guidelines.\n",
        "\n",
        "**5. Dataset Creation**\n",
        "- Include the document, summary pairs, and annotations. Ensure a balanced dataset with diverse examples.\n",
        "- Check for consistent and reliable annotations through cross-validation and reviews.\n",
        "\n",
        "**6. Post-processing**\n",
        "- Normalize scores or rankings for consistency.\n",
        "- Format the dataset to meet the requirements of your training framework.\n",
        "\n",
        "**7. Reward Model Integration**\n",
        "- Initialize Reward Model: Start with the supervised fine-tuned (SFT) summarization model.\n",
        "- Add a Linear Head: Attach a randomly initialized linear layer to the model that outputs a scalar value for scoring summaries.\n",
        "\n",
        "\n",
        "Creating a comparison dataset involves collecting and preprocessing data, generating initial summaries, obtaining human annotations through pairwise comparisons, structuring and formatting the dataset, and performing quality control. Finally, integrate the reward model by initializing it from the SFT model and adding a linear head for evaluation. This dataset and model setup are essential for training in a reinforcement learning from human feedback (RLHF) framework."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "pSf9CV6wliGy"
      },
      "source": [
        "### Question 2 (2.5 Points)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "D467FvoJliGy"
      },
      "source": [
        "**If you have 100 pairs of summaries, and for each pair one summary is prefered, how would you structurre your training data for the reward model?**"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "0JvmNLzCliG1"
      },
      "source": [
        "**Data Structure Definition**\n",
        "\n",
        "Each entry should include:\n",
        "* The document text.\n",
        "* The chosen summary.\n",
        "* The rejected summary.\n",
        "\n",
        "\n",
        "**Data Processing**\n",
        "\n",
        "Pairwise Data Extraction:\n",
        "* Extract pairs of summaries from the dataset.\n",
        "* Assign them to \"chosen\" and \"rejected\" based on the preference.\n",
        "\n",
        "\n",
        "```python\n",
        "# Assume we have a list of comparisons\n",
        "comparisons = [\n",
        "    {\n",
        "        \"prompt\": \"Full text of document 1\",\n",
        "        \"summary_1\": \"Generated summary 1\",\n",
        "        \"summary_2\": \"Generated summary 2\",\n",
        "        \"label\": 0 # Assign wich summary is prefered\n",
        "    },\n",
        "    {\n",
        "        \"prompt\": \"Full text of document 2\",\n",
        "        \"summary_1\": \"Generated summary 3\",\n",
        "        \"summary_2\": \"Generated summary 4\",\n",
        "        \"label\": 1 # Assign wich summary is non-prefered\n",
        "    }\n",
        "]\n",
        "\n",
        "# Prepare data for training\n",
        "train_data = []\n",
        "for comparison in comparisons:\n",
        "    document = comparison[\"prompt\"]\n",
        "    summary_1 = comparison[\"summary_1\"]\n",
        "    summary_2 = comparison[\"summary_2\"]\n",
        "    preferred_summary = comparison[\"label\"]\n",
        "    \n",
        "    if preferred_summary == 0:\n",
        "        train_data.append({\n",
        "            \"prompt\": document,\n",
        "            \"chosen\": summary_1,\n",
        "            \"rejected\": summary_2\n",
        "        })\n",
        "    else:\n",
        "        train_data.append({\n",
        "            \"prompt\": document,\n",
        "            \"chosen\": summary_2,\n",
        "            \"rejected\": summary_1\n",
        "        })\n",
        "\n",
        "```\n",
        "\n",
        "Finally we have this dataset:\n",
        "\n",
        "\n",
        "```json\n",
        "[\n",
        "    {\n",
        "        \"prompt\": \"Full text of document 1\",\n",
        "        \"chosen\": \"prefered summary\",\n",
        "        \"rejected\": \"non prefered summary\"\n",
        "    },\n",
        "    {\n",
        "        \"prompt\": \"Full text of document 2\",\n",
        "        \"chosen\": \"prefered summary\",\n",
        "        \"rejected\": \"non prefered summary\"\n",
        "    }\n",
        "]\n",
        "```"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "n2_7DbzzliG1"
      },
      "source": [
        "### Raw Input"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "SsTnuvmiliG1"
      },
      "source": [
        "Now, we'll create a list of dicts using the create_comparison_dataset function (shown below), where each dict has two keys - chosen and rejected. The value of each key is the prompt (or Reddit post) concatenated with the summary.\n",
        "\n",
        "**Note:** You can increase the number of training examples."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 13,
      "metadata": {
        "id": "yaBQ0g5EliG1"
      },
      "outputs": [],
      "source": [
        "def create_comparison_dataset(\n",
        "     path=\"CarperAI/openai_summarize_comparisons\", split=\"train\"\n",
        " ):\n",
        "     dataset = load_dataset(path, split=split)\n",
        "     if split == \"test\":\n",
        "         dataset = dataset.select(range(1000))\n",
        "     elif split == \"train\":\n",
        "         dataset = dataset.select(range(10000))\n",
        "\n",
        "     pairs = []\n",
        "     for sample in tqdm(dataset):\n",
        "         pair = {}\n",
        "         prompt = sample[\"prompt\"]\n",
        "         chosen_summary = sample[\"chosen\"]\n",
        "         rejected_summary = sample[\"rejected\"]\n",
        "         if chosen_summary == rejected_summary:\n",
        "             continue\n",
        "         if  len(chosen_summary.split()) < 5 or len(rejected_summary.split()) < 5:\n",
        "             continue\n",
        "         pair[\"chosen\"] = prompt + \"\\n\" + chosen_summary\n",
        "         pair[\"rejected\"] = prompt + \"\\n\" + rejected_summary\n",
        "         pairs.append(pair)\n",
        "     return pairs\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "lfiEAWcUliG1"
      },
      "source": [
        "### Pairwise Dataloader (2.5 points)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "WtYfW4QYliG1"
      },
      "source": [
        "The PairwiseDataset class shown below tokenizes the chosen and rejected \"summaries\". The dataset class return the input_ids and attention_masks for both chosen and rejected summaries, in this part you should complete the **PairwiseDataset class.**"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 14,
      "metadata": {
        "id": "0emJGN1hliG1"
      },
      "outputs": [],
      "source": [
        "class PairwiseDataset(Dataset):\n",
        "    def __init__(self, pairs, tokenizer, max_length=512):\n",
        "        self.pairs = pairs\n",
        "        self.tokenizer = tokenizer\n",
        "        self.max_length = max_length\n",
        "\n",
        "    def __len__(self):\n",
        "        return len(self.pairs)\n",
        "\n",
        "    def __getitem__(self, idx):\n",
        "        pair = self.pairs[idx]\n",
        "\n",
        "        chosen_encodings = self.tokenizer(pair[\"chosen\"], truncation=True, max_length=self.max_length, padding=\"max_length\", return_tensors=\"pt\")\n",
        "        rejected_encodings = self.tokenizer(pair[\"rejected\"], truncation=True, max_length=self.max_length, padding=\"max_length\", return_tensors=\"pt\")\n",
        "\n",
        "        return [{\n",
        "            \"chosen_input_ids\": chosen_encodings[\"input_ids\"].squeeze(),\n",
        "            \"chosen_attention_mask\": chosen_encodings[\"attention_mask\"].squeeze(),\n",
        "            \"rejected_input_ids\": rejected_encodings[\"input_ids\"].squeeze(),\n",
        "            \"rejected_attention_mask\": rejected_encodings[\"attention_mask\"].squeeze()\n",
        "        }]"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "wy_jIkb5liG1"
      },
      "source": [
        "### Data Collator (2.5 Points)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ueJlf6e9liG1"
      },
      "source": [
        "The DataCollatorReward class creates batches (dict) of data for our reward model. The collator returns:\n",
        "\n",
        "*   input_ids: collator concatenates the chosen and rejected summaries' input_ids across dim=0.\n",
        "*   attention_mask: collator concatenates the chosen and rejected summaries' attention_mask across dim=0.\n",
        "\n",
        "*   labels: collator creates a tensor of zeros for chosen summaries and a tensor of ones for rejected summaries concatenated across dim=0."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 15,
      "metadata": {
        "id": "6XcjknZhliG2"
      },
      "outputs": [],
      "source": [
        "class DataCollatorReward:\n",
        "    def __call__(self, data):\n",
        "        batch = {}\n",
        "\n",
        "        chosen_input_ids = torch.stack([item[0][\"chosen_input_ids\"] for item in data])\n",
        "        rejected_input_ids = torch.stack([item[0][\"rejected_input_ids\"] for item in data])\n",
        "        \n",
        "        chosen_attention_mask = torch.stack([item[0][\"chosen_attention_mask\"] for item in data])\n",
        "        rejected_attention_mask = torch.stack([item[0][\"rejected_attention_mask\"] for item in data])\n",
        "        \n",
        "        # Concatenate chosen and rejected summaries\n",
        "        batch[\"input_ids\"] = torch.cat((chosen_input_ids, rejected_input_ids), dim=0)\n",
        "        batch[\"attention_mask\"] = torch.cat((chosen_attention_mask, rejected_attention_mask), dim=0)\n",
        "\n",
        "        # Create labels: 0 for chosen summaries, 1 for rejected summaries\n",
        "        chosen_labels = torch.zeros(chosen_input_ids.size(0), dtype=torch.long)\n",
        "        rejected_labels = torch.ones(rejected_input_ids.size(0), dtype=torch.long)\n",
        "        batch[\"labels\"] = torch.cat((chosen_labels, rejected_labels), dim=0)\n",
        "\n",
        "        return batch"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "cF7h0HuHliG2"
      },
      "source": [
        "### What is happening in reward model?"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "2zsVduq1liG2"
      },
      "source": [
        "Here, we have a Reddit post and two summaries (<font color='green'><b>chosen</b></font> and <font color='red'><b>rejected</b></font>) as input.\n",
        "\n",
        "The ground truth label (**labels**) is the human feedback (<font color='green'><b>0 for chosen</b></font> and <font color='red'><b>1 for rejected</b></font>). And the loss function (pairwise ranking loss) is given as:\n",
        "\n",
        "$$\\text{loss}(r_{\\theta}) = -\\mathbb{E}_{(x, y_0, y_1, i) \\sim D} \\left[ \\log \\left( \\sigma \\left( r_{\\theta}(x, y_i) - r_{\\theta}(x, y_{1-i}) \\right) \\right) \\right]\n",
        "$$.\n",
        "\n",
        "\n",
        "where:\n",
        "- $ x $ is the post,\n",
        "- $ y_0 $ and $ y_1 $ are the summaries,\n",
        "- $ i $ in {0, 1} indicates which summary is preferred by humans,\n",
        "- $ r_{\\theta}(x, y) $ is the reward model that returns a scalar value for the post $ x $ and the summary $ y $,\n",
        "- $ \\sigma $ is the sigmoid function.\n",
        "\n",
        "\n",
        "The reward model $ r_{\\theta} $ takes the post $ x $ and the summary $ y $ and returns a scalar value. The value is computed for both summaries and a sigmoid activation is applied to the difference.\n",
        "\n",
        "Finally, the negative log is computed.\n",
        "\n",
        "This loss function encourages the model to give higher scores to human-preferred summaries.\n",
        "\n",
        "**How to code this?**\n",
        "\n",
        "Our model receives input prepared by the data collator.\n",
        "\n",
        "*   This input is passed through the GPT-2 model to get the final hidden states.\n",
        "\n",
        "*   The hidden state is then passed through the linear layer to get a reward score.\n",
        "\n",
        "*   For each batch fed into the model, the first half is the chosen summaries, and the second half is the rejected summaries.\n",
        "\n",
        "*   The forward method of the model iterates through each input sample to compute pairwise loss.\n",
        "*  Return loss and chosen summaries scores and rejected summaries scores."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "n1M63SHQliG2"
      },
      "source": [
        "### Question 3 (2.5 Points)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "pU4AnDFjliG2"
      },
      "source": [
        "**What is the goal of pairwise ranking loss? and how we achieve this goal?**"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "gvQR7PGJliG2"
      },
      "source": [
        "**Goal of Pairwise Ranking Loss**\n",
        "\n",
        "The goal of the pairwise ranking loss in the context of training a reward model is to learn a scoring function $ r_{\\theta} $ that assigns higher scores to preferred outputs compared to less preferred outputs for the same input. In other words, the reward model should learn to rank the preferred (chosen) summaries higher than the rejected ones.\n",
        "\n",
        "**Achieving the Goal**\n",
        "\n",
        "To achieve this goal, the loss function operates as follows:\n",
        "\n",
        "1. **Score Calculation**:\n",
        "   - The reward model $ r_{\\theta} $ computes scores for both the chosen summary $ y_0 $ and the rejected summary $ y_1 $.\n",
        "\n",
        "2. **Score Difference**:\n",
        "   - Calculate the difference in scores: $ r_{\\theta}(x, y_i) - r_{\\theta}(x, y_{1-i}) $. Here, $ y_i $ is the preferred summary, and $ y_{1-i} $ is the less preferred one.\n",
        "\n",
        "3. **Sigmoid Function**:\n",
        "   - Apply the sigmoid function to this difference. The sigmoid function $ \\sigma(z) $ maps the difference to a probability range (0, 1), representing the <ins>likelihood</ins> that the chosen summary $ y_i $ is better than the rejected summary $y_{1-i}$.\n",
        "\n",
        "4. **Log Probability**:\n",
        "   - Take the logarithm of this probability. The log probability penalizes the model more heavily for being confident but wrong.\n",
        "\n",
        "5. **Expectation**:\n",
        "   - The expectation $ \\mathbb{E}_{(x, y_0, y_1, i) \\sim D} $ is taken over all pairs in the dataset, aiming to minimize this expected loss.\n",
        "\n",
        "Now, For each input $x$:\n",
        "1. We generate scores for both chosen and rejected summaries using the reward model.\n",
        "2. Compute the pairwise ranking loss using the formula.\n",
        "3. Backpropagate the loss and update the model parameters $\\theta$ to minimize the loss.\n",
        "\n",
        "=> minimizing the $loss$ => assigns higher scores to preferred outputs and lower scores to non-preferred for the same input."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "vONyXMTzliG2"
      },
      "source": [
        "### Question 4 (2.5 Points)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "8WtZsAmOliG2"
      },
      "source": [
        "**Explain that in the process of training the reward model, how the pairwise ranking loss can avoid the problem of huge score difference between the answers (summaries) and why is this useful?**"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "7acT3rHRliG2"
      },
      "source": [
        "**Avoiding Huge Score Differences:**\n",
        "\n",
        "**1. Sigmoid Function**\n",
        "\n",
        "The sigmoid function $ \\sigma(z) = \\frac{1}{1 + e^{-z}} $ maps the input $ z $ to a value between 0 and 1. This function has a property where:\n",
        "- Small differences between $ r_{\\theta}(x, y_i) $ and $ r_{\\theta}(x, y_{1-i}) $ produce values close to 0.5.\n",
        "- Large positive differences produce values close to 1.\n",
        "- Large negative differences produce values close to 0.\n",
        "\n",
        "**2. Log Probability**\n",
        "\n",
        "Taking the logarithm of the sigmoid output further stabilizes the learning process. The logarithm function compresses the range of values, ensuring that extremely confident predictions (whether correct or incorrect) do not dominate the loss computation excessively. This avoids scenarios where huge score differences could lead to very large or very small gradient updates, which could destabilize training.\n",
        "\n",
        "**Why This Is Useful:**\n",
        "\n",
        "**1. Stable Training**\n",
        "\n",
        "Without this bounding, the gradients could become too large or too small, leading to issues like exploding or vanishing gradients. This stability allows the model to learn more effectively across iterations.\n",
        "\n",
        "**2. Focus on Relative Differences**\n",
        "\n",
        "The pairwise ranking loss focuses on the relative ranking of the chosen and rejected summaries rather than their absolute scores. This is particularly useful because:\n",
        "- It aligns with the goal of ranking preferred summaries higher, regardless of the actual score magnitudes.\n",
        "- It allows the model to learn from the ranking order rather than being affected by absolute score values, which can vary widely.\n",
        "\n",
        "**3. Robustness to Outliers**\n",
        "\n",
        "Outliers or extremely different summaries can skew the training process if absolute score differences are used directly. By focusing on the relative difference and using the sigmoid function, the model becomes more robust to such outliers, preventing them from disproportionately influencing the training."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "NioLnLEZliG2"
      },
      "source": [
        "### Implementing The Reward Model (5 Points)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 20,
      "metadata": {
        "id": "GdyikVCTliG2"
      },
      "outputs": [],
      "source": [
        "class GPTRewardModel(nn.Module):\n",
        "    def __init__(self, model_path):\n",
        "        super().__init__()\n",
        "        self.model = AutoModelForCausalLM.from_pretrained(model_path)\n",
        "        self.config = self.model.config\n",
        "        self.model.lm_head = nn.Linear(self.config.hidden_size, 1)\n",
        "        self.loss_fn = nn.BCEWithLogitsLoss()\n",
        "\n",
        "    def forward(self, input_ids, attention_mask=None, labels=None):\n",
        "        outputs = self.model(input_ids=input_ids, attention_mask=attention_mask)\n",
        "        logits = outputs.logits[:, -1, :]\n",
        "        reward = logits.squeeze(-1)\n",
        "\n",
        "        if labels is not None:\n",
        "            loss = self.loss_fn(reward, labels.float())\n",
        "            return loss, reward\n",
        "\n",
        "        return reward\n",
        "\n",
        "    def save_pretrained(self, save_directory):\n",
        "        self.model.save_pretrained(save_directory)\n",
        "        torch.save(self.state_dict(), f\"{save_directory}/reward_model.bin\")\n",
        "\n",
        "    @classmethod\n",
        "    def from_pretrained(cls, load_directory):\n",
        "        model = cls(load_directory)\n",
        "        model.model = AutoModelForCausalLM.from_pretrained(load_directory)\n",
        "        model.load_state_dict(torch.load(f\"{load_directory}/reward_model.bin\"), strict=False)\n",
        "        return model"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "UOd4qInaliG2"
      },
      "source": [
        "**After finishing the above code, could you please explain how the scores for the selected summaries and the scores for the rejected summaries are calculated in your code?** (2.5 Points)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "PYfEbDKVliG2"
      },
      "source": [
        "The reward model uses the GPT model's capability to transform input texts into meaningful embeddings, averages these embeddings or use the hidden state of first token, and then applies a linear transformation to arrive at a scalar score.\n",
        "\n",
        "This score is intended to reflect the quality or preference of the summary relative to other summaries, based on the training it received. The procedure is identical for both chosen and rejected summaries; what differs is the nature of the text input (i.e., whether the text is from a chosen or rejected summary)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ZIJxF3SrliG3"
      },
      "source": [
        "### Load datasets"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 17,
      "metadata": {
        "id": "PbcvI9_sliG3"
      },
      "outputs": [
        {
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "100%|██████████| 10000/10000 [00:00<00:00, 41315.09it/s]\n",
            "100%|██████████| 1000/1000 [00:00<00:00, 38450.57it/s]\n"
          ]
        }
      ],
      "source": [
        "# Create the comparisons datasets\n",
        "data_path = comparissions_path\n",
        "train_pairs = create_comparison_dataset(data_path, \"train\")\n",
        "val_pairs = create_comparison_dataset(data_path, \"test\")\n",
        "\n",
        "# Make pairwise datasets for training\n",
        "max_length = 550\n",
        "train_dataset = PairwiseDataset(train_pairs, tokenizer, max_length=max_length)\n",
        "val_dataset = PairwiseDataset(val_pairs, tokenizer, max_length=max_length)\n",
        "\n",
        "# Create the collator to gather batches of pairwise comparisons\n",
        "data_collator = DataCollatorReward()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "YhWhVCwHliG3"
      },
      "source": [
        "### Load Model and Tokenizer"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "MAgDrdE5liG3"
      },
      "source": [
        "Initialize the reward model from the SFT GPT-2 model."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 18,
      "metadata": {
        "id": "6ifGgccvliG3"
      },
      "outputs": [],
      "source": [
        "model = GPTRewardModel(CONFIG.output_dir)\n",
        "\n",
        "tokenizer = AutoTokenizer.from_pretrained(CONFIG.model_name)\n",
        "tokenizer.pad_token = tokenizer.eos_token\n",
        "\n",
        "# Freeze the first 70% of the hidden layers of the reward model backbone\n",
        "layers = model.model.transformer.h\n",
        "num_layers = len(layers)\n",
        "num_unfrozen = int(0.3 * num_layers)\n",
        "for layer in layers[:-num_unfrozen]:\n",
        "  layer.requires_grad_(False)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "aTmCQxLaliG3"
      },
      "source": [
        "### Define Compute metric function (2.5 Points)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "JOZSHkSMliG3"
      },
      "source": [
        "In this part you should implement the accuracy of our GPTRewardModel."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 13,
      "metadata": {
        "id": "oMvztXmlliG3"
      },
      "outputs": [],
      "source": [
        "def compute_metrics(eval_preds):\n",
        "    predictions = eval_preds.predictions\n",
        "    batch_size = predictions.shape[0] // 2\n",
        "    \n",
        "    chosen_rewards = predictions[:batch_size]\n",
        "    rejected_rewards = predictions[batch_size:]\n",
        "    \n",
        "    correct_predictions = (chosen_rewards > rejected_rewards).sum().item()\n",
        "    total_predictions = batch_size\n",
        "    \n",
        "    accuracy = correct_predictions / total_predictions\n",
        "    \n",
        "    return {\"accuracy\": accuracy}"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "4lVSVRJcliG3"
      },
      "source": [
        "### Train"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 15,
      "metadata": {
        "id": "USnIpmXdliG3"
      },
      "outputs": [],
      "source": [
        "    training_args = TrainingArguments(\n",
        "      output_dir=\"rm_checkpoint/\",\n",
        "      num_train_epochs=3,\n",
        "      logging_steps=100,\n",
        "      gradient_accumulation_steps=4,\n",
        "      save_strategy=\"steps\",\n",
        "      evaluation_strategy=\"steps\",\n",
        "      per_device_train_batch_size=1,\n",
        "      per_device_eval_batch_size=1,\n",
        "      eval_accumulation_steps=1,\n",
        "      eval_steps=500,\n",
        "      save_steps=1000,\n",
        "      warmup_steps=100,\n",
        "      logging_dir=\"./logs\",\n",
        "      fp16=True,\n",
        "      bf16=False,\n",
        "      learning_rate=1e-5,\n",
        "      save_total_limit=1,\n",
        "    )"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 16,
      "metadata": {
        "id": "nhju2_0wliG3"
      },
      "outputs": [],
      "source": [
        "trainer = Trainer(\n",
        "        model=model,\n",
        "        args=training_args,\n",
        "        train_dataset=train_dataset,\n",
        "        compute_metrics=compute_metrics,\n",
        "        eval_dataset=val_dataset,\n",
        "        data_collator=data_collator,\n",
        "    )"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 17,
      "metadata": {},
      "outputs": [],
      "source": [
        "torch.cuda.empty_cache()"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 18,
      "metadata": {},
      "outputs": [
        {
          "data": {
            "text/html": [
              "\n",
              "    <div>\n",
              "      \n",
              "      <progress value='7500' max='7500' style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
              "      [7500/7500 40:50, Epoch 3/3]\n",
              "    </div>\n",
              "    <table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              " <tr style=\"text-align: left;\">\n",
              "      <th>Step</th>\n",
              "      <th>Training Loss</th>\n",
              "      <th>Validation Loss</th>\n",
              "      <th>Accuracy</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <td>500</td>\n",
              "      <td>0.711800</td>\n",
              "      <td>0.694558</td>\n",
              "      <td>0.541000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>1000</td>\n",
              "      <td>0.699800</td>\n",
              "      <td>0.692472</td>\n",
              "      <td>0.526000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>1500</td>\n",
              "      <td>0.694600</td>\n",
              "      <td>0.688616</td>\n",
              "      <td>0.532000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>2000</td>\n",
              "      <td>0.694900</td>\n",
              "      <td>0.685901</td>\n",
              "      <td>0.532000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>2500</td>\n",
              "      <td>0.694000</td>\n",
              "      <td>0.687172</td>\n",
              "      <td>0.534000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>3000</td>\n",
              "      <td>0.697000</td>\n",
              "      <td>0.687097</td>\n",
              "      <td>0.528000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>3500</td>\n",
              "      <td>0.692000</td>\n",
              "      <td>0.684100</td>\n",
              "      <td>0.528000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>4000</td>\n",
              "      <td>0.684400</td>\n",
              "      <td>0.685113</td>\n",
              "      <td>0.526000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>4500</td>\n",
              "      <td>0.687900</td>\n",
              "      <td>0.683922</td>\n",
              "      <td>0.529000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>5000</td>\n",
              "      <td>0.696700</td>\n",
              "      <td>0.684569</td>\n",
              "      <td>0.521000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>5500</td>\n",
              "      <td>0.691600</td>\n",
              "      <td>0.681460</td>\n",
              "      <td>0.524000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>6000</td>\n",
              "      <td>0.686900</td>\n",
              "      <td>0.680205</td>\n",
              "      <td>0.535000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>6500</td>\n",
              "      <td>0.685000</td>\n",
              "      <td>0.681243</td>\n",
              "      <td>0.529000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>7000</td>\n",
              "      <td>0.685200</td>\n",
              "      <td>0.679623</td>\n",
              "      <td>0.527000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>7500</td>\n",
              "      <td>0.681100</td>\n",
              "      <td>0.680011</td>\n",
              "      <td>0.531000</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table><p>"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        }
      ],
      "source": [
        "trainer.train()\n",
        "trainer.save_model(CONFIG.output_dir_rm)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 19,
      "metadata": {},
      "outputs": [
        {
          "data": {
            "text/html": [
              "\n",
              "    <div>\n",
              "      \n",
              "      <progress value='1000' max='1000' style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
              "      [1000/1000 00:21]\n",
              "    </div>\n",
              "    "
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        },
        {
          "data": {
            "text/plain": [
              "{'eval_loss': 0.6800112724304199,\n",
              " 'eval_accuracy': 0.531,\n",
              " 'eval_runtime': 21.1465,\n",
              " 'eval_samples_per_second': 47.289,\n",
              " 'eval_steps_per_second': 47.289,\n",
              " 'epoch': 3.0}"
            ]
          },
          "execution_count": 19,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "trainer.evaluate()"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 20,
      "metadata": {},
      "outputs": [],
      "source": [
        "model.save_pretrained('rm_model')"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "l8X8eRJDliG3"
      },
      "source": [
        "## Section Three: PPO Fine Tuning (25 Points)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "odye-6kMliG3"
      },
      "source": [
        "### Question 5 (2.5 points)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "nznG1XPUliG3"
      },
      "source": [
        "**What is PPO algorithm? and how it works?**"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "nwnofdGaliG3"
      },
      "source": [
        "**What is PPO algorithm?**\n",
        "\n",
        "Proximal Policy Optimization (PPO) is a RL algorithm developed by OpenAI, designed to be simpler to implement and more stable than earlier methods such as TRPO. PPO has become one of the most popular algorithms in the field of reinforcement learning due to its effectiveness and ease of use.\n",
        "\n",
        "PPO is a **policy-based** method, meaning it directly parameterizes the policy (i.e., the action-selection strategy) and optimizes it. This is in contrast to value-based methods like Q-learning, which optimize a value function and derive the policy from it. PPO is designed to provide stable policy updates by limiting the changes to the policy during training.\n",
        "\n",
        "**How does PPO work?**\n",
        "\n",
        "1. **Collect Trajectories**: Gather data by running the current policy in the environment.\n",
        "\n",
        "2. **Compute Advantages**: Estimate the advantage of actions using collected data.\n",
        "\n",
        "3. **Update Policy**: Optimize the policy using a <ins>clipped objective</ins> function to prevent large updates.\n",
        "\n",
        "4. **Update Value Function**: Adjust the value function to better predict future rewards.\n",
        "\n",
        "5. **Repeat**: Continue the cycle with more data collection and updates."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "HbN18nGpliG3"
      },
      "source": [
        "### Question 6 (2.5 points)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "051cbYOHliG3"
      },
      "source": [
        "**Is the PPO algorithm an on-policy or off-policy reinforcement learning algorithm? Explain its functionality within our context (RLHF).**"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "tnNg_n7wliG3"
      },
      "source": [
        "Proximal Policy Optimization (PPO) is an on-policy reinforcement learning algorithm. This means it updates the policy using data collected from the current policy, rather than from a different policy (as in off-policy methods).\n",
        "\n",
        "It is well-suited for RLHF because it directly utilizes the human feedback collected based on the agent’s current policy. This ensures that the feedback is relevant and immediately applicable, facilitating stable and consistent policy improvements."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Qf52CnPNliG3"
      },
      "source": [
        "### Question 7 (2.5 points)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "z7ULCFnCliG3"
      },
      "source": [
        "**Imagine a mini-batch of data has arrived, and we want to optimize the policy of generating summaries to maximize the reward using gradient ascent.**\n",
        "\n",
        "**Why shouldn't this policy change too much and, in other words, become overoptimized? (Answer based on the respond you provided to the previous question.)**\n",
        "\n",
        "**What do they do to solve this problem?**"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "BRntFnNyliG4"
      },
      "source": [
        "**Why Shouldn't the Policy Change Too Much?**\n",
        "\n",
        "* **Stability:** Large changes can make the policy unstable, leading to poor performance as it may not generalize well.\n",
        "\n",
        "* **Overfitting:** Rapid optimization can overfit to the current mini-batch, reducing performance on new data.\n",
        "\n",
        "* **Exploration vs. Exploitation:** Drastic changes can prematurely converge to suboptimal actions, reducing exploration.\n",
        "\n",
        "**How PPO solves this problem?**\n",
        "\n",
        "- **Clipping Mechanism**: PPO uses a clipped objective function to limit the extent of policy updates. The key idea is to limit how much the probability ratio $ r_t(\\theta) = \\frac{\\pi_\\theta(a_t | s_t)}{\\pi_{\\theta_{\\text{old}}}(a_t | s_t)} $ can deviate from 1. So we have this term $\\text{clip}(r_t(\\theta), 1 - \\epsilon, 1 + \\epsilon)$.\n",
        "\n",
        "    (Note: $ \\epsilon $ is Small threshold that controls the size of updates)\n",
        "\n",
        "- **Surrogate Objective**: The surrogate objective balances policy improvement with stability. It ensures that policy updates are significant only if they provide a clear advantage without deviating excessively from the current policy.\n",
        "\n",
        "  $  L_\\text{CLIP}(\\theta) = \\mathbb{E}_t \\left[ \\min \\left( r_t(\\theta) \\hat{A}_t, \\text{clip}(r_t(\\theta), 1 - \\epsilon, 1 + \\epsilon) \\hat{A}_t \\right) \\right] $\n",
        "\n",
        "- **Entropy Bonus**: An entropy term is added to the objective to maintain diversity in action selection, preventing premature convergence to a narrow set of actions."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "zqXq1dD1liG4"
      },
      "source": [
        "### Question 8 (2.5 points)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "r9DxQa18liG4"
      },
      "source": [
        "**What is the overestimation problem in the ppo fine-tuning? and why it happens?**"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "mTB67tQfliG4"
      },
      "source": [
        "The overestimation problem occurs when the PPO algorithm tends to overvalue certain actions, leading to biased and suboptimal decisions.\n",
        "\n",
        "**Why it happens?**\n",
        "\n",
        "1. **High Reward Variance:** Noisy rewards can lead to inaccurate advantage estimates, causing the policy to overvalue some actions due to random fluctuations.\n",
        "\n",
        "2. **Positive Bias:** The advantage function can introduce a bias, making the policy favor actions that seem better than they are.\n",
        "\n",
        "3. **Clipping Mechanism:** While it stabilizes updates, the clipping mechanism doesn't eliminate existing biases, which can persist and accumulate.\n",
        "\n",
        "4. **Insufficient Exploration:** During fine-tuning, the policy might not explore enough new actions, reinforcing initial overestimations."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "yspjLgeVliG4"
      },
      "source": [
        "### Question 9 (2.5 points)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "66q19irOliG4"
      },
      "source": [
        "**What potential issue could arise when aligning a language model with human values? What solution has been proposed to address this issue?**"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "0UMTxiDdliG4"
      },
      "source": [
        "\\# WRITE YOUR ANSWER HERE\n",
        ".\n",
        ".\n",
        "."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "5oFLh6auliG4"
      },
      "source": [
        "### Question 10 (2.5 points)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ppgwgfp7liG4"
      },
      "source": [
        "**We know that the objective function of the ppo tuning is as follows:**\n",
        "\n",
        "$$ \\text{objective}(\\phi) = \\mathbb{E}_{(x,y) \\sim D_{\\pi_{\\phi}^{\\text{RL}}}} \\left[ r_{\\theta}(x, y) - \\beta \\log \\left( \\frac{\\pi_{\\phi}^{\\text{RL}}(y \\mid x)}{\\pi^{\\text{SFT}}(y \\mid x)} \\right) \\right] + \\gamma \\mathbb{E}_{x \\sim D_{\\text{pretrain}}} \\left[ \\log(\\pi_{\\phi}^{\\text{RL}}(x)) \\right]\n",
        " $$\n",
        "\n",
        "**In the above objective function, the differentiation is with respect to $Φ$, yet the term $r_{\\theta}(x, y)$ appears in the function.**\n",
        "\n",
        "**Explain why $r$ appears in the objective function despite being a function of $\\theta$ and why its derivative is not zero.**"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "MszsZK-fliG4"
      },
      "source": [
        "\\# WRITE YOUR ANSWER HERE"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "gJgxRLkDliG4"
      },
      "source": [
        "### Question 11 (5 points)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "wnkwyraDliG4"
      },
      "source": [
        "**Another term present in the objective function is $\\beta \\log \\left( \\frac{\\pi_{\\phi}^{\\text{RL}}(y \\mid x)}{\\pi^{\\text{SFT}}(y \\mid x)} \\right)$, which is the KLD between the initial policy distribution and the policy being learned. Explain why this term cannot be calculated directly.**\n",
        "\n",
        "\n",
        "**For a detailed explanation of how the KLD is calculated in this context, please read this <br>[blog post](http://joschu.net/blog/kl-approx.html). Afterward, provide an explanation of the calculation process.**"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Bn9aDtEtliG4"
      },
      "source": [
        "\\# WRITE YOUR ANSWER HERE"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "8h1VdPVJliG4"
      },
      "source": [
        "### Question 12 (5 points)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "6yVF745CliG4"
      },
      "source": [
        "**How does DPO improve fine tuning?**\n",
        "\n",
        "**In DPO, our loss optimizing function to optimize for the policy is:**\n",
        "$$\\text{L}_\\text{DPO}(\\pi_\\theta; \\pi_\\text{ref}) = -\\mathbb{E}_{(x,y_w,y_l) \\sim D}\n",
        "\\left[\n",
        "\\log \\sigma\n",
        "\\left(\n",
        "\\beta \\log \\frac{\\pi_\\theta(y_w | x)}{\\pi_\\text{ref}(y_w | x)}\n",
        "- \\beta \\log \\frac{\\pi_\\theta(y_l | x)}{\\pi_\\text{ref}(y_l | x)}\n",
        "\\right)\n",
        "\\right]$$\n",
        "\n",
        "**Please explain the terms of this loss function and interpret the function.**\n",
        "\n",
        "\n",
        "**Now, please calculate the gradient of this loss function and Intuitively explain each term in the gradient of loss function.**"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "OsGwADFOliG4"
      },
      "source": [
        "\\# WRITE YOUR ANSWER HERE"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "KgMiQEouliG4"
      },
      "source": [
        "### Run PPO Fine Tuning (optional)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "GNoIe9weliG4"
      },
      "source": [
        "Because of the limitations of Google Colab, If you have access to an extra  GPU you can run below code for ppo fine tuning."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "4MDEVHxGliG5"
      },
      "outputs": [],
      "source": [
        "%pip install -U git+https://github.com/CarperAI/trlx.git"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 8,
      "metadata": {
        "id": "RgklnHR3liG5"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "[2024-06-16 15:49:44,851] [INFO] [real_accelerator.py:203:get_accelerator] Setting ds_accelerator to cuda (auto detect)\n",
            "\u001b[93m [WARNING] \u001b[0m async_io requires the dev libaio .so object and headers but these were not found.\n"
          ]
        },
        {
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "df: /teamspace/studios/this_studio/.triton/autotune: No such file or directory\n",
            "/home/zeus/miniconda3/envs/cloudspace/compiler_compat/ld: cannot find -laio: No such file or directory\n",
            "collect2: error: ld returned 1 exit status\n"
          ]
        },
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\u001b[93m [WARNING] \u001b[0m async_io: please install the libaio-dev package with apt\n",
            "\u001b[93m [WARNING] \u001b[0m If libaio is already installed (perhaps from source), try setting the CFLAGS and LDFLAGS environment variables to where it can be found.\n",
            "\u001b[93m [WARNING] \u001b[0m Please specify the CUTLASS repo directory as environment variable $CUTLASS_PATH\n",
            "\u001b[93m [WARNING] \u001b[0m sparse_attn requires a torch version >= 1.5 and < 2.0 but detected 2.2\n",
            "\u001b[93m [WARNING] \u001b[0m using untested triton version (2.2.0), only 1.0.0 is known to be compatible\n"
          ]
        }
      ],
      "source": [
        "import torch\n",
        "from datasets import load_dataset\n",
        "from tqdm import tqdm\n",
        "from transformers import AutoTokenizer\n",
        "\n",
        "import trlx\n",
        "from trlx.data.configs import (\n",
        "    ModelConfig,\n",
        "    OptimizerConfig,\n",
        "    SchedulerConfig,\n",
        "    TokenizerConfig,\n",
        "    TrainConfig,\n",
        "    TRLConfig,\n",
        ")\n",
        "from trlx.models.modeling_ppo import PPOConfig"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 9,
      "metadata": {
        "id": "J5S_0bM9liG5"
      },
      "outputs": [],
      "source": [
        "config = TRLConfig(\n",
        "    train=TrainConfig(\n",
        "        seq_length=550,\n",
        "        epochs=50,\n",
        "        total_steps=100000,\n",
        "        batch_size=4,\n",
        "        checkpoint_interval=10000,\n",
        "        eval_interval=200,\n",
        "        pipeline=\"PromptPipeline\",\n",
        "        trainer=\"AcceleratePPOTrainer\",\n",
        "    ),\n",
        "    model=ModelConfig(\n",
        "        model_path=\"gpt2\",\n",
        "        num_layers_unfrozen=8,\n",
        "    ),\n",
        "    tokenizer=TokenizerConfig(\n",
        "        tokenizer_path=\"gpt2\",\n",
        "        truncation_side=\"right\",\n",
        "    ),\n",
        "    optimizer=OptimizerConfig(\n",
        "        name=\"adamw\",\n",
        "        kwargs={\n",
        "            \"lr\": 5.0e-6,\n",
        "            \"betas\": [0.9, 0.999],\n",
        "            \"eps\": 1.0e-8,\n",
        "            \"weight_decay\": 0.01,\n",
        "        },\n",
        "    ),\n",
        "    scheduler=SchedulerConfig(\n",
        "        name=\"cosine_annealing\",\n",
        "        kwargs={\n",
        "            \"T_max\": 100000,\n",
        "            \"eta_min\": 5.0e-6,\n",
        "        },\n",
        "    ),\n",
        "    method=PPOConfig(\n",
        "        name=\"PPOConfig\",\n",
        "        num_rollouts=128,\n",
        "        chunk_size=16,\n",
        "        ppo_epochs=4,\n",
        "        init_kl_coef=0.1,\n",
        "        target=6,\n",
        "        horizon=10000,\n",
        "        gamma=1,\n",
        "        lam=0.95,\n",
        "        cliprange=0.2,\n",
        "        cliprange_value=0.2,\n",
        "        vf_coef=0.2,\n",
        "        scale_reward=None,\n",
        "        ref_mean=None,\n",
        "        ref_std=None,\n",
        "        cliprange_reward=10,\n",
        "        gen_kwargs={\n",
        "            \"max_new_tokens\": 50,\n",
        "        },\n",
        "    ),\n",
        ")\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 40,
      "metadata": {
        "id": "7kDLLWSGliG5"
      },
      "outputs": [],
      "source": [
        "REWARD_CHECKPOINT_PATH= CONFIG.output_dir_rm + '/training_args.bin' #TODO\n",
        "SFT_MODEL_PATH = CONFIG.output_dir #TODO"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 24,
      "metadata": {
        "id": "O0e5L7T2liG5"
      },
      "outputs": [],
      "source": [
        "from typing import List\n",
        "def get_scores(samples: List[str]):\n",
        "  scores_list = []\n",
        "  batch_size = 2\n",
        "  for i in range(0, len(samples), batch_size):\n",
        "    sub_samples = samples[i : i + batch_size]\n",
        "    sub_samples = [\"<|startoftext|>\" + chosen + \"<|endoftext|>\" for chosen in sub_samples]\n",
        "    encodings_dict = rw_tokenizer(\n",
        "        sub_samples,\n",
        "        truncation=True,\n",
        "        max_length=config.train.seq_length,\n",
        "        padding=\"max_length\",\n",
        "        return_tensors=\"pt\",\n",
        "        )\n",
        "    input_ids = encodings_dict[\"input_ids\"].to(rw_device)\n",
        "    attn_masks = encodings_dict[\"attention_mask\"].to(rw_device)\n",
        "    input_ids = input_ids.repeat(2, 1)\n",
        "    attn_masks = attn_masks.repeat(2, 1)\n",
        "    with torch.no_grad():\n",
        "      sub_scores = rw_model(input_ids=input_ids, attention_mask=attn_masks)\n",
        "    scores_list.append(sub_scores[\"chosen_end_scores\"])\n",
        "  scores = torch.cat(scores_list, dim=0)\n",
        "  return scores"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 35,
      "metadata": {
        "id": "4BeXGRuyliG5"
      },
      "outputs": [],
      "source": [
        "rw_tokenizer = AutoTokenizer.from_pretrained(\"gpt2\")\n",
        "rw_tokenizer.pad_token = rw_tokenizer.eos_token\n",
        "rw_model = GPTRewardModel(SFT_MODEL_PATH)\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 39,
      "metadata": {},
      "outputs": [
        {
          "data": {
            "text/plain": [
              "'rm_checkpoint'"
            ]
          },
          "execution_count": 39,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "CONFIG.output_dir_rm + ''"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 42,
      "metadata": {},
      "outputs": [
        {
          "ename": "UnpicklingError",
          "evalue": "invalid load key, '>'.",
          "output_type": "error",
          "traceback": [
            "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
            "\u001b[0;31mUnpicklingError\u001b[0m                           Traceback (most recent call last)",
            "Cell \u001b[0;32mIn[42], line 1\u001b[0m\n\u001b[0;32m----> 1\u001b[0m rw_model\u001b[38;5;241m.\u001b[39mload_state_dict(\u001b[43mtorch\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mload\u001b[49m\u001b[43m(\u001b[49m\u001b[43mCONFIG\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43moutput_dir_rm\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m+\u001b[39;49m\u001b[43m \u001b[49m\u001b[38;5;124;43m'\u001b[39;49m\u001b[38;5;124;43m/model.safetensors\u001b[39;49m\u001b[38;5;124;43m'\u001b[39;49m\u001b[43m)\u001b[49m, strict\u001b[38;5;241m=\u001b[39m\u001b[38;5;28;01mFalse\u001b[39;00m)\n\u001b[1;32m      2\u001b[0m rw_model\u001b[38;5;241m.\u001b[39mhalf()\n\u001b[1;32m      3\u001b[0m rw_model\u001b[38;5;241m.\u001b[39meval()\n",
            "File \u001b[0;32m/home/zeus/miniconda3/envs/cloudspace/lib/python3.10/site-packages/torch/serialization.py:1040\u001b[0m, in \u001b[0;36mload\u001b[0;34m(f, map_location, pickle_module, weights_only, mmap, **pickle_load_args)\u001b[0m\n\u001b[1;32m   1038\u001b[0m     \u001b[38;5;28;01mexcept\u001b[39;00m \u001b[38;5;167;01mRuntimeError\u001b[39;00m \u001b[38;5;28;01mas\u001b[39;00m e:\n\u001b[1;32m   1039\u001b[0m         \u001b[38;5;28;01mraise\u001b[39;00m pickle\u001b[38;5;241m.\u001b[39mUnpicklingError(UNSAFE_MESSAGE \u001b[38;5;241m+\u001b[39m \u001b[38;5;28mstr\u001b[39m(e)) \u001b[38;5;28;01mfrom\u001b[39;00m \u001b[38;5;28;01mNone\u001b[39;00m\n\u001b[0;32m-> 1040\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43m_legacy_load\u001b[49m\u001b[43m(\u001b[49m\u001b[43mopened_file\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mmap_location\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mpickle_module\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mpickle_load_args\u001b[49m\u001b[43m)\u001b[49m\n",
            "File \u001b[0;32m/home/zeus/miniconda3/envs/cloudspace/lib/python3.10/site-packages/torch/serialization.py:1258\u001b[0m, in \u001b[0;36m_legacy_load\u001b[0;34m(f, map_location, pickle_module, **pickle_load_args)\u001b[0m\n\u001b[1;32m   1252\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m \u001b[38;5;28mhasattr\u001b[39m(f, \u001b[38;5;124m'\u001b[39m\u001b[38;5;124mreadinto\u001b[39m\u001b[38;5;124m'\u001b[39m) \u001b[38;5;129;01mand\u001b[39;00m (\u001b[38;5;241m3\u001b[39m, \u001b[38;5;241m8\u001b[39m, \u001b[38;5;241m0\u001b[39m) \u001b[38;5;241m<\u001b[39m\u001b[38;5;241m=\u001b[39m sys\u001b[38;5;241m.\u001b[39mversion_info \u001b[38;5;241m<\u001b[39m (\u001b[38;5;241m3\u001b[39m, \u001b[38;5;241m8\u001b[39m, \u001b[38;5;241m2\u001b[39m):\n\u001b[1;32m   1253\u001b[0m     \u001b[38;5;28;01mraise\u001b[39;00m \u001b[38;5;167;01mRuntimeError\u001b[39;00m(\n\u001b[1;32m   1254\u001b[0m         \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mtorch.load does not work with file-like objects that do not implement readinto on Python 3.8.0 and 3.8.1. \u001b[39m\u001b[38;5;124m\"\u001b[39m\n\u001b[1;32m   1255\u001b[0m         \u001b[38;5;124mf\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mReceived object of type \u001b[39m\u001b[38;5;130;01m\\\"\u001b[39;00m\u001b[38;5;132;01m{\u001b[39;00m\u001b[38;5;28mtype\u001b[39m(f)\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;130;01m\\\"\u001b[39;00m\u001b[38;5;124m. Please update to Python 3.8.2 or newer to restore this \u001b[39m\u001b[38;5;124m\"\u001b[39m\n\u001b[1;32m   1256\u001b[0m         \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mfunctionality.\u001b[39m\u001b[38;5;124m\"\u001b[39m)\n\u001b[0;32m-> 1258\u001b[0m magic_number \u001b[38;5;241m=\u001b[39m \u001b[43mpickle_module\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mload\u001b[49m\u001b[43m(\u001b[49m\u001b[43mf\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mpickle_load_args\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m   1259\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m magic_number \u001b[38;5;241m!=\u001b[39m MAGIC_NUMBER:\n\u001b[1;32m   1260\u001b[0m     \u001b[38;5;28;01mraise\u001b[39;00m \u001b[38;5;167;01mRuntimeError\u001b[39;00m(\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mInvalid magic number; corrupt file?\u001b[39m\u001b[38;5;124m\"\u001b[39m)\n",
            "\u001b[0;31mUnpicklingError\u001b[0m: invalid load key, '>'."
          ]
        }
      ],
      "source": [
        "rw_model.load_state_dict(torch.load(CONFIG.output_dir_rm + '/training_args.bin'), strict=False)\n",
        "rw_model.half()\n",
        "rw_model.eval()\n",
        "rw_model.to(CONFIG.device)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "lHL-Bs1eliG5"
      },
      "outputs": [],
      "source": [
        "def get_prompt_dataset(prompts, max_length):\n",
        "  formatted_prompts = []\n",
        "  for i in tqdm(range(len(prompts))):\n",
        "    tmp = tokenizer.decode(\n",
        "        tokenizer(\n",
        "              prompts[i].split(\"TL;DR:\")[0],\n",
        "              truncation=True,\n",
        "              max_length=max_length - 5,\n",
        "              add_special_tokens=False,\n",
        "            )[\"input_ids\"],\n",
        "            skip_special_tokens=True,\n",
        "            ).strip()\n",
        "    tmp = tmp + \"\\nTL;DR:\"\n",
        "    tmp = tokenizer.decode(\n",
        "          tokenizer(tmp, truncation=True, max_length=max_length, add_special_tokens=False)[\"input_ids\"],\n",
        "          skip_special_tokens=True,\n",
        "        ).strip()\n",
        "    formatted_prompts.append(tmp)\n",
        "  return formatted_prompts"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "D3NqMZu5liG5"
      },
      "outputs": [],
      "source": [
        "def reward_fn(samples: List[str], **kwargs):\n",
        "  original_samples = [text.split(\"TL;DR:\")[0] + \"TL;DR: \" for text in samples]\n",
        "  original_samples = [text + post_summary_dict[text.strip()] for text in original_samples]\n",
        "  original_scores = get_scores(original_samples)\n",
        "  scores = get_scores(samples)\n",
        "  norms_scores = scores - original_scores\n",
        "  return norms_scores"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "_a8fbDBuliG5"
      },
      "outputs": [],
      "source": [
        "tokenizer = AutoTokenizer.from_pretrained(config.tokenizer.tokenizer_path)\n",
        "tokenizer.pad_token = tokenizer.eos_token\n",
        "tokenizer.padding_side = \"left\"\n",
        "max_length_input = config.train.seq_length - config.method.gen_kwargs[\"max_new_tokens\"]\n",
        "\n",
        "# Load the dataset\n",
        "dataset = load_dataset(\"carperai/openai_summarize_tldr\")\n",
        "\n",
        "# Get the total number of samples in the training dataset\n",
        "total_samples = len(dataset[\"train\"])\n",
        "\n",
        "# Generate random indices for selecting samples\n",
        "random_indices = random.sample(range(total_samples), 5000)\n",
        "\n",
        "# Select samples using random indices\n",
        "train_set = [(dataset[\"train\"][index][\"prompt\"], dataset[\"train\"][index][\"label\"]) for index in random_indices]\n",
        "val_set = [(sample[\"prompt\"], sample[\"label\"]) for sample in dataset[\"valid\"]]\n",
        "\n",
        "# Split contents into summaries and labels\n",
        "train_posts, train_summaries = zip(*train_set)\n",
        "val_posts, val_summaries = zip(*val_set)\n",
        "\n",
        "# Get the OpenAI summaries\n",
        "post_summary_dict = {}\n",
        "train_prompts = get_prompt_dataset(train_posts, max_length_input)\n",
        "for i in range(len(train_prompts)):\n",
        "  post_summary_dict[train_prompts[i]] = train_summaries[i]\n",
        "val_prompts = get_prompt_dataset(val_posts, max_length_input)\n",
        "for i in range(len(val_prompts)):\n",
        "  post_summary_dict[val_prompts[i]] = val_summaries[i]"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "GugzrPa_liG5"
      },
      "outputs": [],
      "source": [
        "trainer = trlx.train(\n",
        "    reward_fn=reward_fn,\n",
        "    prompts=train_prompts,\n",
        "    eval_prompts=val_prompts[0:1000],  # sampling 1000 validation prompts for evaluation speed in training\n",
        "    config=config,\n",
        "  )"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "wBP-hkzEliG5"
      },
      "source": [
        "# Quantization (26 points)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "jR4LIwnCliG5"
      },
      "source": [
        "Quantization is a technique used to reduce the precision of neural network weights and activations, typically from floating-point to a lower-bit representation, such as 8-bit or 4-bit integers. The primary goal of quantization is to reduce the memory footprint and computational requirements of deep learning models, enabling the loading of larger models that would normally not fit into available memory, and speeding up the inference process."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "yX3IbcrVliG5"
      },
      "source": [
        "## A simple example (2 points)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "rtNa3uKRliG5"
      },
      "source": [
        "Let's see what happens when a we quantize a 32-bit floating-point number."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 2,
      "metadata": {
        "id": "s1jaB5hsliG5"
      },
      "outputs": [],
      "source": [
        "# Import neccesary libraries\n",
        "import numpy as np\n",
        "import matplotlib.pyplot as plt"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "oxo4FpTFliG5"
      },
      "source": [
        "Defining two functions which responsible for quantizing and dequantizing the input number:"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 3,
      "metadata": {
        "id": "N7t0v2RxliG6"
      },
      "outputs": [],
      "source": [
        "def quantize(value, num_bits=4):\n",
        "    quantized_value = np.round(value * (2**(num_bits - 1) - 1))\n",
        "    return int(quantized_value)\n",
        "\n",
        "def dequantize(quantized_value, num_bits=4):\n",
        "    value = quantized_value / (2**(num_bits - 1) - 1)\n",
        "    return float(value)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Kuf-64O5liG6"
      },
      "source": [
        "Consider the value `0.326`, the quantized values in 4 and 8 bits are:"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 27,
      "metadata": {
        "id": "bOQ-MJETliG6"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "4-bit: 2\n",
            "8-bit: 41\n"
          ]
        }
      ],
      "source": [
        "q_4bit = quantize(value=0.326, num_bits=4)\n",
        "q_8bit = quantize(value=0.326, num_bits=8)\n",
        "\n",
        "print(f'4-bit: {q_4bit}')\n",
        "print(f'8-bit: {q_8bit}')"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "kSFVYdbFliG6"
      },
      "source": [
        "And if we dequantize it to original full precision values we would have:"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 28,
      "metadata": {
        "id": "cXgpYWgyliG6"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "4-bit: 0.2857142857142857\n",
            "8-bit: 0.3228346456692913\n"
          ]
        }
      ],
      "source": [
        "print(f'4-bit: {dequantize(quantized_value=q_4bit, num_bits=4)}')\n",
        "print(f'8-bit: {dequantize(quantized_value=q_8bit, num_bits=8)}')"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "62EqHz2FliG6"
      },
      "source": [
        "8-bit quantization preserves the original precision with very little degradationa and 4-bit quantization does incur more precision loss, but the level of loss can still be tolerated for many applications."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "mKjkbccyliG6"
      },
      "source": [
        "To understand the precision loss from 4-bit and 8-bit quantization, plot the function $y = x^2$ in the range of $[-1, 1]$, and compare the original values to the values obtained after quantization and dequantization for both 4-bit and 8-bit cases. (2 points)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 4,
      "metadata": {
        "id": "s0BMw9LzliG6"
      },
      "outputs": [],
      "source": [
        "def func(x):\n",
        "    return x**2\n",
        "\n",
        "def quantized_func(x, num_bits=4):\n",
        "    return dequantize(quantize(func(x), num_bits), num_bits)\n",
        "\n",
        "x_values = np.linspace(-1, 1, 1000)\n",
        "y_original = func(x_values)\n",
        "\n",
        "y_q_4bit = [quantized_func(x, num_bits=4) for x in x_values]\n",
        "y_q_8bit = [quantized_func(x, num_bits=8) for x in x_values]"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 99,
      "metadata": {},
      "outputs": [
        {
          "data": {
            "image/png": "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",
            "text/plain": [
              "<Figure size 900x600 with 1 Axes>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        }
      ],
      "source": [
        "plt.figure(figsize=(9, 6))\n",
        "\n",
        "# Plot the original function\n",
        "plt.plot(x_values, y_original, label='Original Function: $y = x^2$', linewidth=3)\n",
        "plt.plot(x_values, y_q_8bit, label='8-bit Quantization', linestyle='--')\n",
        "plt.plot(x_values, y_q_4bit, label='4-bit Quantization', linestyle='--')\n",
        "\n",
        "plt.title('Effect of Quantization on $y = x^2$')\n",
        "plt.xlabel('x')\n",
        "plt.ylabel('y')\n",
        "plt.legend()\n",
        "plt.grid(True)\n",
        "plt.show()\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "6ASecpj7liG6"
      },
      "source": [
        "## QLoRA: Efficient Finetuning of Quantized LLMs (24 points)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Akr6OhtWliG6"
      },
      "source": [
        "Now, it's time to see the magic of quantization. We are going to fine-tune Mistral 7B model based on the method proposed by [QLoRA: Efficient Finetuning of Quantized LLMs](https://arxiv.org/abs/2305.14314)."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Az1kOYjOliG6"
      },
      "source": [
        "### Prerequisite"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ARgU36wzliG6"
      },
      "source": [
        "Installing and importing libraries which we will need later."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "ITqClfSxliG6"
      },
      "outputs": [],
      "source": [
        "%pip install -q accelerate==0.29.3\n",
        "%pip install -q bitsandbytes==0.43.1\n",
        "%pip install -q trl==0.8.6\n",
        "%pip install -q peft==0.10.0\n",
        "%pip install -q transformers==4.40.0"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 2,
      "metadata": {
        "id": "DKvCF2-cliG6"
      },
      "outputs": [],
      "source": [
        "from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig\n",
        "from trl import SFTTrainer, DataCollatorForCompletionOnlyLM\n",
        "import torch"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "4tNWeHkDliG6"
      },
      "source": [
        "### Hugging Face Login"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "zkpuugsVliG6"
      },
      "source": [
        "For some language models, you need to agree to share your contact information to access the model. Mistral 7B is one of them. The steps you should take are as follows:\n",
        "\n",
        "1.   Create a Hugging Face account if you don't have one.\n",
        "2.   From Settings > Access Tokens, generate a new token.\n",
        "3.   From [this link](https://huggingface.co/mistralai/Mistral-7B-v0.1) agree to access the repository.\n",
        "\n",
        "Now, run the code below to login to your account."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "25vTyCA-liG6"
      },
      "outputs": [],
      "source": [
        "# from huggingface_hub import login\n",
        "# access_token_read = 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\n",
        "# login(token = access_token_read)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 3,
      "metadata": {},
      "outputs": [
        {
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "e182ddc6f67d437bb4a0c4b049242f80",
              "version_major": 2,
              "version_minor": 0
            },
            "text/plain": [
              "VBox(children=(HTML(value='<center> <img\\nsrc=https://huggingface.co/front/assets/huggingface_logo-noborder.sv…"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        }
      ],
      "source": [
        "from huggingface_hub import notebook_login\n",
        "notebook_login()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "8lKc4I9vliG6"
      },
      "source": [
        "### The Model (6 points)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "8C71CI8ZliG6"
      },
      "source": [
        "To load the model, first we need a quantization configuration to set how we want to quantize the model. We are going to use `BitsAndBytesConfig` to achieve our goal. `BitsAndBytesConfig` is the easiest option for quantizing a model to 8 and 4-bit.\n",
        "\n",
        "Define the quantization configuration in the cell below based on the fact that we are looking for 4-bit quantization. Also, QLoRA paper proposed two techniques: 4-bit NormalFloat(NF4) quantization and Double Quantization. Set these two setup in our configuration too. To speedup the computation try to set computational type as bf16. (2 points)\n",
        "\n",
        "Try to explain what is the Double Quantization technique. You can use the provided paper to read about it. (4 points)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "qmKYvFTlliG6"
      },
      "source": [
        "The Double Quantization technique, as proposed in the QLoRA paper, involves two stages of quantization:\n",
        "\n",
        "* **First Stage (Low Precision):** The model is quantized to a low bit-width (e.g., 4-bit) using techniques like NF4 (4-bit NormalFloat) quantization for activations and a similar low-precision scheme for weights (e.g., bfloat16).\n",
        "\n",
        "* **Second Stage (Higher Precision):** After training the model with low-precision quantization, the weights are re-quantized to a higher precision (e.g., 8-bit or higher). This step helps mitigate the performance degradation caused by the initial aggressive quantization, potentially improving accuracy while still benefiting from the computational efficiency of low-precision inference.\n",
        "\n",
        "The key idea behind Double Quantization is to leverage the benefits of both low-precision and higher-precision quantization methods: the former for efficient computation and the latter for maintaining or improving model accuracy. This two-stage approach can be particularly effective in scenarios where maintaining high accuracy is critical despite using low-precision hardware or constraints."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 4,
      "metadata": {
        "id": "LiU0JmvkliG6"
      },
      "outputs": [],
      "source": [
        "model_name = 'mistralai/Mistral-7B-v0.1'\n",
        "\n",
        "quantization_config = BitsAndBytesConfig(\n",
        "    load_in_4bit=True,\n",
        "    bnb_4bit_quant_type=\"nf4\",\n",
        "    bnb_4bit_compute_dtype=torch.float16,\n",
        "    bnb_4bit_use_double_quant=True,\n",
        "    )"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "JB6GGQ9wliG7"
      },
      "source": [
        "Now, let's load the model."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 10,
      "metadata": {
        "id": "CMg4--m2liG7"
      },
      "outputs": [
        {
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "b67edc62a3754ff2b461153fc7343b9a",
              "version_major": 2,
              "version_minor": 0
            },
            "text/plain": [
              "model.safetensors.index.json:   0%|          | 0.00/25.1k [00:00<?, ?B/s]"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        },
        {
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "1b8340f407c64af2aa99baee713afa52",
              "version_major": 2,
              "version_minor": 0
            },
            "text/plain": [
              "Downloading shards:   0%|          | 0/2 [00:00<?, ?it/s]"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        },
        {
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "3c01ef825f264c0c8423a58f9dd82c47",
              "version_major": 2,
              "version_minor": 0
            },
            "text/plain": [
              "model-00001-of-00002.safetensors:   0%|          | 0.00/9.94G [00:00<?, ?B/s]"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        },
        {
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "6ece7471c77c4d018fbf721ee7b2cdb1",
              "version_major": 2,
              "version_minor": 0
            },
            "text/plain": [
              "model-00002-of-00002.safetensors:   0%|          | 0.00/4.54G [00:00<?, ?B/s]"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        },
        {
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "d1d86f103bde46c3bb8a5b1fc901bf4e",
              "version_major": 2,
              "version_minor": 0
            },
            "text/plain": [
              "Loading checkpoint shards:   0%|          | 0/2 [00:00<?, ?it/s]"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        },
        {
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "9330282ab2a84bf9bd5492cf45dd9c07",
              "version_major": 2,
              "version_minor": 0
            },
            "text/plain": [
              "generation_config.json:   0%|          | 0.00/116 [00:00<?, ?B/s]"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        }
      ],
      "source": [
        "# device_map = {\"\": 0}\n",
        "device_map = \"auto\"\n",
        "\n",
        "model = AutoModelForCausalLM.from_pretrained(\n",
        "    model_name,\n",
        "    quantization_config=quantization_config,\n",
        "    device_map=device_map,\n",
        "    use_cache = False,\n",
        "    )"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 6,
      "metadata": {},
      "outputs": [],
      "source": [
        "from peft import prepare_model_for_kbit_training\n",
        "model = prepare_model_for_kbit_training(model)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ZgoxRNz4liG7"
      },
      "source": [
        "Now, we have the quantized model on the memory. You can try to load the model without quantization into memory to see if it is possible."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 7,
      "metadata": {
        "id": "9f-h69h3liG7"
      },
      "outputs": [],
      "source": [
        "tokenizer = AutoTokenizer.from_pretrained(model_name)\n",
        "tokenizer.pad_token = tokenizer.eos_token\n",
        "\n",
        "model.config.pad_token_id = tokenizer.pad_token_id\n",
        "model.config.end_token_id = tokenizer.eos_token_id"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ufcvIS94liG7"
      },
      "source": [
        "### Generate text using the pre-trained model (1 point)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "0HoJK7uzliG7"
      },
      "source": [
        "In this part we are going to test the pre-trained model to see its capabilities."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 27,
      "metadata": {
        "id": "USJG6n_JliG7"
      },
      "outputs": [],
      "source": [
        "# Function which generate outputs corresponding to input prompts\n",
        "def generate_output(model, inputs, max_length=50):\n",
        "\n",
        "    tokenized_inputs = tokenizer(inputs, padding=True, return_tensors=\"pt\").to('cuda')\n",
        "    output = model.generate(\n",
        "        input_ids=tokenized_inputs[\"input_ids\"],\n",
        "        attention_mask=tokenized_inputs[\"attention_mask\"],\n",
        "        max_new_tokens=max_length,\n",
        "        repetition_penalty=1.5,\n",
        "        early_stopping=False,\n",
        "        eos_token_id=tokenizer.eos_token_id,\n",
        "        pad_token_id=tokenizer.pad_token_id,\n",
        "    )\n",
        "\n",
        "    text = tokenizer.batch_decode(output, skip_special_tokens=True)\n",
        "    return text"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "S-4aZeJ0liG7"
      },
      "source": [
        "Using the `generate_output` function try to produce output of the pre-trained model for the following prompts: (1 point)\n",
        "\n",
        "* \"What's up?\"\n",
        "* 'Hello, How are you?'\n",
        "* 'hello, Can you help me?'\n",
        "* 'what is the capital of France?'\n",
        "* 'Hi! what do you do for a living?'\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 17,
      "metadata": {
        "id": "ZzmwARwNliG7"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "[1] What's up?\n",
            "I am a 20 year old student from the Netherlands. I have been playing guitar for about five years now and started to play bass in my band last summer (we are called 'The Dirty Pigs'). We just released our first\n",
            "\n",
            "[2] Hello, How are you?\n",
            "I am a newbie in the world of Arduino and I have some questions. Firstly let me introduce myself: my name is Andrea (23 years old) from Italy; i'm an Electronic Engineer student at University\n",
            "\n",
            "[3] hello, Can you help me?\n",
            "I have a problem with my 2013 MacBook Pro. I bought it in December and since then the battery has been draining very quickly (about an hour of use). The first time this happened was when i had to charge\n",
            "\n",
            "[4] what is the capital of France?\n",
            "\n",
            "Paris. Paris, city and commune (1982 pop., 350,476), prefecture of Seine-et-Marne dept., Île-deFrance region, N central France\n",
            "\n",
            "[5] Hi! what do you do for a living?\n",
            "I am an artist and I work in the field of art, design & technology. My main focus is on interactive installations that are based around human interaction with light or sound as well as video projections onto objects/sculptures etc…\n",
            "\n"
          ]
        }
      ],
      "source": [
        "sample_prompts = [\"What's up?\", 'Hello, How are you?', 'hello, Can you help me?', 'what is the capital of France?', 'Hi! what do you do for a living?']\n",
        "\n",
        "for i, input in enumerate(sample_prompts):\n",
        "    text = generate_output(model, input)\n",
        "    print(f'[{i+1}] {text[0]}\\n')"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "rf0dg5nBliG7"
      },
      "source": [
        "### The Dataset (4 points)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "uUiem55MliG7"
      },
      "source": [
        "We loaded a 7B LLM into the memory and succeeded to inference from the model. Now, using [this dataset](https://huggingface.co/datasets/SAGI-1/Greetings_DPO_dataset_V1) we want to fine-tune our pre-trained model.\n",
        "\n",
        "After loading the dataset, choose 75 samples of it. We would like to backpropagate only the tokens of the completion and not the prompt itself. In order to do this, use `DataCollatorForCompletionOnlyLM` (check the examples provided in [this link](https://huggingface.co/docs/trl/en/sft_trainer#train-on-completions-only)). Also, you can ignore the `rejected` column."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 37,
      "metadata": {},
      "outputs": [
        {
          "ename": "AttributeError",
          "evalue": "'DatasetDict' object has no attribute 'select'",
          "output_type": "error",
          "traceback": [
            "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
            "\u001b[0;31mAttributeError\u001b[0m                            Traceback (most recent call last)",
            "Cell \u001b[0;32mIn[37], line 3\u001b[0m\n\u001b[1;32m      1\u001b[0m dataset_name \u001b[38;5;241m=\u001b[39m \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mSAGI-1/Greetings_DPO_dataset_V1\u001b[39m\u001b[38;5;124m\"\u001b[39m\n\u001b[1;32m      2\u001b[0m dataset \u001b[38;5;241m=\u001b[39m load_dataset(dataset_name)\n\u001b[0;32m----> 3\u001b[0m selected_train_dataset \u001b[38;5;241m=\u001b[39m \u001b[43mdataset\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mselect\u001b[49m(\u001b[38;5;28mrange\u001b[39m(\u001b[38;5;241m750\u001b[39m))\n\u001b[1;32m      4\u001b[0m selected_test_dataset \u001b[38;5;241m=\u001b[39m dataset\u001b[38;5;241m.\u001b[39mselect(\u001b[38;5;28mrange\u001b[39m(\u001b[38;5;241m75\u001b[39m))\n",
            "\u001b[0;31mAttributeError\u001b[0m: 'DatasetDict' object has no attribute 'select'"
          ]
        }
      ],
      "source": [
        "dataset_name = \"SAGI-1/Greetings_DPO_dataset_V1\"\n",
        "dataset = load_dataset(dataset_name)\n",
        "selected_train_dataset = dataset.select(range(750))\n",
        "selected_test_dataset = dataset.select(range(75))"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 29,
      "metadata": {},
      "outputs": [],
      "source": [
        "def formatting_prompts_func(example):\n",
        "    output_texts = []\n",
        "    for i in range(len(example['prompt'])):\n",
        "        text = f\"### Question: {example['prompt'][i]}\\n ### Answer: {example['chosen'][i]}\"\n",
        "        output_texts.append(text)\n",
        "    return output_texts\n",
        "\n",
        "response_template = \" ### Answer:\"\n",
        "collator = DataCollatorForCompletionOnlyLM(response_template, tokenizer=tokenizer)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "fGmRA7I7liG7"
      },
      "source": [
        "### Fine-tuning (8 points)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ZBXpgSDZliG7"
      },
      "source": [
        "We want to fine-tune the model with the help of [LoRA](https://arxiv.org/abs/2106.09685). Try to create a LoRA configuration with rank and alpha parameter both equal to 12. (3 points)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 11,
      "metadata": {
        "id": "KNa8fk8MliG7"
      },
      "outputs": [],
      "source": [
        "import peft\n",
        "from peft import LoraConfig, get_peft_model\n",
        "\n",
        "target_modules = [\"q_proj\", \"v_proj\"]\n",
        "\n",
        "lora_config = peft_config = LoraConfig(\n",
        "    r=12,\n",
        "    lora_alpha=12,\n",
        "    lora_dropout=0.05,\n",
        "    bias=\"none\",\n",
        "    task_type=\"CAUSAL_LM\",\n",
        "    target_modules=target_modules\n",
        ")\n",
        "\n",
        "model = get_peft_model(model, lora_config)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 12,
      "metadata": {},
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "trainable params: 5,111,808 || all params: 7,246,843,904 || trainable%: 0.070538403582537\n"
          ]
        }
      ],
      "source": [
        "model.print_trainable_parameters()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "gR92T3V1liG7"
      },
      "source": [
        "We are going to use SFTtrainer from trl library. In order to work within memory constraints, try to set the batch size and the number of update steps to accumulate the gradients equal to 1. Also, don't forget to enable the `gradient_checkpointing`. (5 points)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 33,
      "metadata": {
        "id": "WxteG8s_liG7"
      },
      "outputs": [],
      "source": [
        "training_args = TrainingArguments(\n",
        "    output_dir='./qlora_results',\n",
        "    num_train_epochs=1,\n",
        "    per_device_train_batch_size=1,\n",
        "    gradient_accumulation_steps=1,\n",
        "    # fp16=True,  # Enable mixed precision training if supported\n",
        "    # bf16=True,  # Enable bf16 training if supported (for faster training on supported hardware)\n",
        "    logging_dir='./qlora_logs',\n",
        "    logging_steps=100,\n",
        "    evaluation_strategy=\"steps\",\n",
        "    eval_steps=100,\n",
        "    save_steps=500,\n",
        "    gradient_checkpointing=True,  # Enable gradient checkpointing\n",
        "    report_to=\"none\",\n",
        ")\n",
        "\n",
        "trainer = SFTTrainer(\n",
        "    model=model,\n",
        "    args=training_args,\n",
        "    train_dataset=dataset,\n",
        "    formatting_func=formatting_prompts_func,\n",
        "    data_collator=collator,\n",
        ")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 34,
      "metadata": {},
      "outputs": [],
      "source": [
        "torch.cuda.empty_cache()"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 35,
      "metadata": {},
      "outputs": [
        {
          "data": {
            "text/html": [
              "\n",
              "    <div>\n",
              "      \n",
              "      <progress value='101' max='891' style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
              "      [101/891 03:09 < 25:16, 0.52 it/s, Epoch 0.11/1]\n",
              "    </div>\n",
              "    <table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              " <tr style=\"text-align: left;\">\n",
              "      <th>Step</th>\n",
              "      <th>Training Loss</th>\n",
              "      <th>Validation Loss</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "  </tbody>\n",
              "</table><p>"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        },
        {
          "ename": "ValueError",
          "evalue": "Trainer: evaluation requires an eval_dataset.",
          "output_type": "error",
          "traceback": [
            "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
            "\u001b[0;31mValueError\u001b[0m                                Traceback (most recent call last)",
            "Cell \u001b[0;32mIn[35], line 1\u001b[0m\n\u001b[0;32m----> 1\u001b[0m trainer_result \u001b[38;5;241m=\u001b[39m \u001b[43mtrainer\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mtrain\u001b[49m\u001b[43m(\u001b[49m\u001b[43m)\u001b[49m\n",
            "File \u001b[0;32m/home/zeus/miniconda3/envs/cloudspace/lib/python3.10/site-packages/trl/trainer/sft_trainer.py:361\u001b[0m, in \u001b[0;36mSFTTrainer.train\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m    358\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mneftune_noise_alpha \u001b[38;5;129;01mis\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m \u001b[38;5;28;01mNone\u001b[39;00m \u001b[38;5;129;01mand\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_trainer_supports_neftune:\n\u001b[1;32m    359\u001b[0m     \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mmodel \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_trl_activate_neftune(\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mmodel)\n\u001b[0;32m--> 361\u001b[0m output \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;43msuper\u001b[39;49m\u001b[43m(\u001b[49m\u001b[43m)\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mtrain\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    363\u001b[0m \u001b[38;5;66;03m# After training we make sure to retrieve back the original forward pass method\u001b[39;00m\n\u001b[1;32m    364\u001b[0m \u001b[38;5;66;03m# for the embedding layer by removing the forward post hook.\u001b[39;00m\n\u001b[1;32m    365\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mneftune_noise_alpha \u001b[38;5;129;01mis\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m \u001b[38;5;28;01mNone\u001b[39;00m \u001b[38;5;129;01mand\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_trainer_supports_neftune:\n",
            "File \u001b[0;32m/home/zeus/miniconda3/envs/cloudspace/lib/python3.10/site-packages/transformers/trainer.py:1859\u001b[0m, in \u001b[0;36mTrainer.train\u001b[0;34m(self, resume_from_checkpoint, trial, ignore_keys_for_eval, **kwargs)\u001b[0m\n\u001b[1;32m   1857\u001b[0m         hf_hub_utils\u001b[38;5;241m.\u001b[39menable_progress_bars()\n\u001b[1;32m   1858\u001b[0m \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[0;32m-> 1859\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43minner_training_loop\u001b[49m\u001b[43m(\u001b[49m\n\u001b[1;32m   1860\u001b[0m \u001b[43m        \u001b[49m\u001b[43margs\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   1861\u001b[0m \u001b[43m        \u001b[49m\u001b[43mresume_from_checkpoint\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mresume_from_checkpoint\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   1862\u001b[0m \u001b[43m        \u001b[49m\u001b[43mtrial\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mtrial\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   1863\u001b[0m \u001b[43m        \u001b[49m\u001b[43mignore_keys_for_eval\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mignore_keys_for_eval\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   1864\u001b[0m \u001b[43m    \u001b[49m\u001b[43m)\u001b[49m\n",
            "File \u001b[0;32m/home/zeus/miniconda3/envs/cloudspace/lib/python3.10/site-packages/transformers/trainer.py:2278\u001b[0m, in \u001b[0;36mTrainer._inner_training_loop\u001b[0;34m(self, batch_size, args, resume_from_checkpoint, trial, ignore_keys_for_eval)\u001b[0m\n\u001b[1;32m   2275\u001b[0m     \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mstate\u001b[38;5;241m.\u001b[39mepoch \u001b[38;5;241m=\u001b[39m epoch \u001b[38;5;241m+\u001b[39m (step \u001b[38;5;241m+\u001b[39m \u001b[38;5;241m1\u001b[39m \u001b[38;5;241m+\u001b[39m steps_skipped) \u001b[38;5;241m/\u001b[39m steps_in_epoch\n\u001b[1;32m   2276\u001b[0m     \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mcontrol \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mcallback_handler\u001b[38;5;241m.\u001b[39mon_step_end(args, \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mstate, \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mcontrol)\n\u001b[0;32m-> 2278\u001b[0m     \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_maybe_log_save_evaluate\u001b[49m\u001b[43m(\u001b[49m\u001b[43mtr_loss\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mgrad_norm\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mmodel\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mtrial\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mepoch\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mignore_keys_for_eval\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m   2279\u001b[0m \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[1;32m   2280\u001b[0m     \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mcontrol \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mcallback_handler\u001b[38;5;241m.\u001b[39mon_substep_end(args, \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mstate, \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mcontrol)\n",
            "File \u001b[0;32m/home/zeus/miniconda3/envs/cloudspace/lib/python3.10/site-packages/transformers/trainer.py:2662\u001b[0m, in \u001b[0;36mTrainer._maybe_log_save_evaluate\u001b[0;34m(self, tr_loss, grad_norm, model, trial, epoch, ignore_keys_for_eval)\u001b[0m\n\u001b[1;32m   2660\u001b[0m metrics \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mNone\u001b[39;00m\n\u001b[1;32m   2661\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mcontrol\u001b[38;5;241m.\u001b[39mshould_evaluate:\n\u001b[0;32m-> 2662\u001b[0m     metrics \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mevaluate\u001b[49m\u001b[43m(\u001b[49m\u001b[43mignore_keys\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mignore_keys_for_eval\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m   2663\u001b[0m     \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_report_to_hp_search(trial, \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mstate\u001b[38;5;241m.\u001b[39mglobal_step, metrics)\n\u001b[1;32m   2665\u001b[0m     \u001b[38;5;66;03m# Run delayed LR scheduler now that metrics are populated\u001b[39;00m\n",
            "File \u001b[0;32m/home/zeus/miniconda3/envs/cloudspace/lib/python3.10/site-packages/transformers/trainer.py:3460\u001b[0m, in \u001b[0;36mTrainer.evaluate\u001b[0;34m(self, eval_dataset, ignore_keys, metric_key_prefix)\u001b[0m\n\u001b[1;32m   3457\u001b[0m \u001b[38;5;66;03m# memory metrics - must set up as early as possible\u001b[39;00m\n\u001b[1;32m   3458\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_memory_tracker\u001b[38;5;241m.\u001b[39mstart()\n\u001b[0;32m-> 3460\u001b[0m eval_dataloader \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mget_eval_dataloader\u001b[49m\u001b[43m(\u001b[49m\u001b[43meval_dataset\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m   3461\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mis_fsdp_xla_v2_enabled:\n\u001b[1;32m   3462\u001b[0m     eval_dataloader \u001b[38;5;241m=\u001b[39m tpu_spmd_dataloader(eval_dataloader)\n",
            "File \u001b[0;32m/home/zeus/miniconda3/envs/cloudspace/lib/python3.10/site-packages/transformers/trainer.py:915\u001b[0m, in \u001b[0;36mTrainer.get_eval_dataloader\u001b[0;34m(self, eval_dataset)\u001b[0m\n\u001b[1;32m    904\u001b[0m \u001b[38;5;250m\u001b[39m\u001b[38;5;124;03m\"\"\"\u001b[39;00m\n\u001b[1;32m    905\u001b[0m \u001b[38;5;124;03mReturns the evaluation [`~torch.utils.data.DataLoader`].\u001b[39;00m\n\u001b[1;32m    906\u001b[0m \n\u001b[0;32m   (...)\u001b[0m\n\u001b[1;32m    912\u001b[0m \u001b[38;5;124;03m        by the `model.forward()` method are automatically removed. It must implement `__len__`.\u001b[39;00m\n\u001b[1;32m    913\u001b[0m \u001b[38;5;124;03m\"\"\"\u001b[39;00m\n\u001b[1;32m    914\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m eval_dataset \u001b[38;5;129;01mis\u001b[39;00m \u001b[38;5;28;01mNone\u001b[39;00m \u001b[38;5;129;01mand\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39meval_dataset \u001b[38;5;129;01mis\u001b[39;00m \u001b[38;5;28;01mNone\u001b[39;00m:\n\u001b[0;32m--> 915\u001b[0m     \u001b[38;5;28;01mraise\u001b[39;00m \u001b[38;5;167;01mValueError\u001b[39;00m(\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mTrainer: evaluation requires an eval_dataset.\u001b[39m\u001b[38;5;124m\"\u001b[39m)\n\u001b[1;32m    917\u001b[0m \u001b[38;5;66;03m# If we have persistent workers, don't do a fork bomb especially as eval datasets\u001b[39;00m\n\u001b[1;32m    918\u001b[0m \u001b[38;5;66;03m# don't change during training\u001b[39;00m\n\u001b[1;32m    919\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;28mhasattr\u001b[39m(\u001b[38;5;28mself\u001b[39m, \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124m_eval_dataloader\u001b[39m\u001b[38;5;124m\"\u001b[39m) \u001b[38;5;129;01mand\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39margs\u001b[38;5;241m.\u001b[39mdataloader_persistent_workers:\n",
            "\u001b[0;31mValueError\u001b[0m: Trainer: evaluation requires an eval_dataset."
          ]
        }
      ],
      "source": [
        "trainer_result = trainer.train()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "bP3XIco2liG7"
      },
      "source": [
        "### Check the fine-tuned model (5 points)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "XtRCMb9ZliG7"
      },
      "source": [
        "Try to check the fine-tuned model on the five aformentioned prompts which we test earlier. Compare you results of fine-tuned and pre-trained model."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "XTdXI7GaliG7"
      },
      "source": [
        "\\# WRITE YOUR ANSWER HERE"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "RTz-Gm0dliG7"
      },
      "outputs": [],
      "source": [
        "# WRITE YOUR CODE HERE"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "XcU69fllliG8"
      },
      "source": [
        "In case you have memory problem try `del` statement for the variables you don't need. The code below also could be beneficial."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "GVJw7WZ6liG8"
      },
      "outputs": [],
      "source": [
        "import gc\n",
        "torch.cuda.empty_cache()\n",
        "gc.collect()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "54Wa42NtliG8"
      },
      "source": [
        "# Instruct Tuning (4 points)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "6DIF-4qNliG8"
      },
      "source": [
        "What we done earlier was somehow instruct tuning in which we fine-tune the pre-trained model to follow the instructions.However, with 75 samples we can't do anything big. As you probably know, the pre-training objective for auto-regressive models is to simply predict the next word. Therefore, they are not optimized to answer to your prompts, rather they try continue the text you provided.\n",
        "\n",
        "In many practical settings (like chatbots) we need the model to follow the instructions provided by the user. So, huge datasets consisting of prompts and their corresponding answer are used to fine-tune these models for interactive use cases.\n",
        "\n",
        "Using the quantization technique we used earlier try to load the quantized version of [Mistral-7B-Instruct](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1) which have been optimized for following the instructions. Check the five samples we used earlier and compare your results.\n",
        "\n",
        "Try to check the provided link to see how you should format the prompt to leverage instruction fine-tuning."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "04HY3MbuliG8"
      },
      "outputs": [],
      "source": [
        "instruct_model = 'mistralai/Mistral-7B-Instruct-v0.1'\n",
        "# WRITE YOUR CODE HERE"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "enKlQTAVliG8"
      },
      "source": [
        "\\# WRITE YOUR ANSWER HERE"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "bbVo3DQQliG8"
      },
      "source": [
        "# Evaluation (25 points)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "SvIAojNGliG8"
      },
      "source": [
        "## Evaluating text using a language model"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "BjIBCESmliG8"
      },
      "source": [
        "One way of evaluating text generation is by using a language model.\n",
        "In this assignment, we want to use BERTScore to compare the similarity of sentences.\n",
        "After reading [BERTScore paper](https://arxiv.org/abs/1904.09675) answer the following questions."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "zux9coTYliG8"
      },
      "source": [
        "### 1)\n",
        "How does BERTScore evaluate semantic equivalence and how is it better than n-gram based metrics like BLEU? (5 points)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "FZ_O4mjnliG8"
      },
      "source": [
        "\\# WRITE YOUR ANSWER HERE"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "bQA5q_fWliG8"
      },
      "source": [
        "### 2)\n",
        "Implement BERTScore recall $R_{BERT}$ in the paper. You don't need to bother with the Importance Weighting. We use a more modern model called deberta which is better than bert in many ways. Then visualize similarity matrix. You can use the visualization code given to you. (7 points)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "HqIkNBZrliG8"
      },
      "outputs": [],
      "source": [
        "import torch\n",
        "import numpy as np\n",
        "from transformers import AutoTokenizer, AutoModel\n",
        "\n",
        "tokenizer = AutoTokenizer.from_pretrained(\"microsoft/deberta-v3-small\")\n",
        "model = AutoModel.from_pretrained(\"microsoft/deberta-v3-small\")\n",
        "\n",
        "def cosine_similarity(a, b):\n",
        "    return # WRITE YOUR CODE HERE (implement using pytorch CosineSimilarity)\n",
        "\n",
        "def bert_score(reference, candidate):\n",
        "\n",
        "\n",
        "    # WRITE YOUR CODE HERE (Tokenize the input text)\n",
        "    ref_tokens = \"tokenize reference sentence\"\n",
        "    candidate_tokens = \"you should know what to do here\"\n",
        "\n",
        "    model.eval()\n",
        "    with torch.no_grad():\n",
        "        # WRITE YOUR CODE HERE (get the embeddings please)\n",
        "        pass\n",
        "\n",
        "\n",
        "    similarities = \"The similarities matrix\"\n",
        "\n",
        "    R_BERT = \"\"\"You can see max in the BERTScore formula in the paper. Then what are you waiting for? Implement the thing.\n",
        "            Pay attention to the fact that you are computing recall. It's important to be careful about\n",
        "              which axis you are getting the max on.\"\"\"\n",
        "    bertscore = R_BERT.mean()\n",
        "\n",
        "\n",
        "    return bertscore, similarities\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "AMHYsxrJliG8"
      },
      "outputs": [],
      "source": [
        "reference = \"Cats are cute\"\n",
        "candidate = \"Cats are annoying\"\n",
        "\n",
        "bertscore, sim_matrix = bert_score(\n",
        "    reference, candidate)\n",
        "\n",
        "print(bertscore)\n",
        "print(sim_matrix)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "PFt2JSZFliG8"
      },
      "outputs": [],
      "source": [
        "import matplotlib.pyplot as plt\n",
        "import numpy as np\n",
        "\n",
        "\n",
        "def heatmap(\n",
        "    matrix,\n",
        "    hide_spines=False,\n",
        "    hide_ticks=False,\n",
        "    figsize=None,\n",
        "    cmap=None,\n",
        "    colorbar=True,\n",
        "    row_names=None,\n",
        "    column_names=None,\n",
        "    column_name_rotation=45,\n",
        "    cell_values=True,\n",
        "    cell_fmt=\".2f\",\n",
        "    cell_font_size=None,\n",
        "):\n",
        "\n",
        "    if row_names is not None and len(row_names) != matrix.shape[0]:\n",
        "        raise AssertionError(\n",
        "            f\"len(row_names) (got {len(row_names)})\"\n",
        "            \" should be equal to number of\"\n",
        "            \" rows in the input \"\n",
        "            f\" array (expect {matrix.shape[0]}).\"\n",
        "        )\n",
        "\n",
        "    if column_names is not None and len(column_names) != matrix.shape[1]:\n",
        "        raise AssertionError(\n",
        "            f\"len(column_names)\"\n",
        "            \" (got {len(column_names)})\"\n",
        "            \" should be equal to number of\"\n",
        "            \" columns in the\"\n",
        "            f\" input array (expect {matrix.shape[1]}).\"\n",
        "        )\n",
        "\n",
        "    fig, ax = plt.subplots(figsize=figsize)\n",
        "    ax.grid(False)\n",
        "\n",
        "    if cmap is None:\n",
        "        cmap = plt.cm.viridis\n",
        "\n",
        "    if figsize is None:\n",
        "        figsize = (len(matrix) * 1.5, len(matrix) * 1.5)\n",
        "\n",
        "    matshow = ax.matshow(matrix, cmap=cmap)\n",
        "\n",
        "    if colorbar:\n",
        "        fig.colorbar(matshow)\n",
        "\n",
        "    normed_matrix = matrix.astype(\"float\") / matrix.max()\n",
        "\n",
        "    if cell_values:\n",
        "        for i in range(matrix.shape[0]):\n",
        "            for j in range(matrix.shape[1]):\n",
        "                cell_text = format(matrix[i, j], cell_fmt)\n",
        "\n",
        "                ax.text(\n",
        "                    x=j,\n",
        "                    y=i,\n",
        "                    size=cell_font_size,\n",
        "                    s=cell_text,\n",
        "                    va=\"center\",\n",
        "                    ha=\"center\",\n",
        "                    color=\"black\"\n",
        "                    if normed_matrix[i, j] > np.max(normed_matrix) / 2\n",
        "                    else \"black\",\n",
        "                )\n",
        "\n",
        "    if row_names is not None:\n",
        "        tick_marks = np.arange(len(row_names))\n",
        "        plt.yticks(tick_marks, row_names)\n",
        "\n",
        "    if column_names is not None:\n",
        "        tick_marks = np.arange(len(column_names))\n",
        "\n",
        "        if column_name_rotation:\n",
        "            plt.xticks(\n",
        "                tick_marks,\n",
        "                column_names,\n",
        "                rotation=column_name_rotation,\n",
        "                ha=\"right\",\n",
        "                rotation_mode=\"anchor\",\n",
        "            )\n",
        "        else:\n",
        "            plt.xticks(tick_marks, column_names)\n",
        "\n",
        "    if hide_spines:\n",
        "        ax.spines[\"right\"].set_visible(False)\n",
        "        ax.spines[\"top\"].set_visible(False)\n",
        "        ax.spines[\"left\"].set_visible(False)\n",
        "        ax.spines[\"bottom\"].set_visible(False)\n",
        "    ax.yaxis.set_ticks_position(\"left\")\n",
        "    ax.xaxis.set_ticks_position(\"bottom\")\n",
        "    if hide_ticks:\n",
        "        ax.axes.get_yaxis().set_ticks([])\n",
        "        ax.axes.get_xaxis().set_ticks([])\n",
        "\n",
        "    return fig, ax"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "N4STcfpIliG8"
      },
      "outputs": [],
      "source": [
        "from matplotlib import cm\n",
        "from matplotlib.colors import ListedColormap\n",
        "\n",
        "viridisBig = cm.get_cmap('Blues', 512)\n",
        "newcmp = ListedColormap(viridisBig(np.linspace(0.0, 0.5, 256)))\n",
        "\n",
        "fig, ax = heatmap(sim_matrix, column_names=reference.split(), row_names=candidate.split(), cmap=newcmp, figsize=(4, 4))\n",
        "\n",
        "plt.tight_layout();\n",
        "plt.show()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "7e0wMXpTliG8"
      },
      "source": [
        "### 3)\n",
        "Using the [Official BERTScore](https://github.com/Tiiiger/bert_score) package. Compare your implementation with the official one and tell us why they differ. (3 points)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "pJQNnr5uliG9"
      },
      "outputs": [],
      "source": [
        "# pip install bert-score"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "v0Sf7hGTliG9"
      },
      "outputs": [],
      "source": [
        "import torch\n",
        "from bert_score import score\n",
        "\n",
        "reference = \"Cats are cute\"\n",
        "candidate = \"Cats are annoying\"\n",
        "\n",
        "P, R, F1 = score(\n",
        "    # WRITE YOUR CODE HERE (complete yourself)\n",
        ")\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "BaAYAIieliG9"
      },
      "source": [
        "### 4)\n",
        "Download datasets \"sst2\" and \"imdb\" from huggingface datasets. Both of these datasets are for the same task but their distribution differs. In what ways their distribution differ according to [this paper](https://aclanthology.org/2021.emnlp-main.835/)?\n",
        "\n",
        "randomly choose ten sentences from negative and ten sentences from positive labels of each dataset (forty sentences in total)\n",
        "\n",
        "use your implementation of BERTScore and compute similarity of\n",
        "\n",
        "a) sentences that are in the same dataset and have same label and take average\n",
        "\n",
        "b) sentences that are in the same dataset and have opposite label and take average\n",
        "\n",
        "c) sentences that are in different datasets and have same label and take average\n",
        "\n",
        "d) sentences that are in the different dataset and have opposite label and take average\n",
        "\n",
        "How these numbers differ and why? (for example for part a you should put each sentence as reference once and compare it with ten candidate sentences including itself, after doing this for all ten sentences you get a hundred scores and then an average.) (10 points)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "jQ5_XWVWliG9"
      },
      "source": [
        "\\# WRITE YOUR ANSWER HERE"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "# AI Assistant"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "* How to create a comparison dataset for a text summarization task for reward model in RLHF?\n",
        "* What is the goal of the pairwise ranking loss in the context of training a reward model?\n",
        "* How the pairwise ranking loss can avoid the problem of huge score difference between the answers (summaries) during training and why is this useful?\n",
        "* What is PPO algorithm? and how it works?\n",
        "* Is the PPO algorithm an on-policy or off-policy reinforcement learning algorithm?\n",
        "* What is the overestimation problem in the ppo fine-tuning? and why it happens?"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": []
    }
  ],
  "metadata": {
    "colab": {
      "provenance": [],
      "toc_visible": true
    },
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    },
    "language_info": {
      "name": "python"
    },
    "widgets": {
      "application/vnd.jupyter.widget-state+json": {
        "005b8f782bdb4ab2a8a5e40a75674d96": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "04588e43dc174ae5bcebc0ef01d581dc": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "FloatProgressModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "FloatProgressModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "ProgressView",
            "bar_style": "success",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_58cabc7c155747fc8cbe2f27f1beb289",
            "max": 26,
            "min": 0,
            "orientation": "horizontal",
            "style": "IPY_MODEL_cce757f48382449da0d3e3bf87272284",
            "value": 26
          }
        },
        "054c5f4fc64744a7a46e991855c7b386": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HTMLModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_d18ab3331f1a42f1af89aca50b9e78e6",
            "placeholder": "​",
            "style": "IPY_MODEL_08a9f29c73b54e0cbc89738191edb1a8",
            "value": "config.json: 100%"
          }
        },
        "08a9f29c73b54e0cbc89738191edb1a8": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "0b831cc698eb4838960af0add2becea1": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "0e017c1b69fe4e89b3a5e169c9387445": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "0faf1d3472974ceca6eeda7e9d5f7906": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "FloatProgressModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "FloatProgressModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "ProgressView",
            "bar_style": "success",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_c6c5a944a23e456ebb6c69805e4d4e2e",
            "max": 532,
            "min": 0,
            "orientation": "horizontal",
            "style": "IPY_MODEL_196db38e52bf47e0a529d725ad71d007",
            "value": 532
          }
        },
        "12ce1164d18249b3807e5ec45aac4a84": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "16f10c9d3be140829e228ad4b3aa7705": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "18439c99c16d4162b8a2a52d86dc5847": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "196db38e52bf47e0a529d725ad71d007": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "ProgressStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "ProgressStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "bar_color": null,
            "description_width": ""
          }
        },
        "19a4391701ad4399887e35757ed7eea6": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "1dd9d180974942d89cabcc4d318a6cf0": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "1f99bdeb99dc43f882494cb9a8d1f664": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "ProgressStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "ProgressStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "bar_color": null,
            "description_width": ""
          }
        },
        "2053f4c66e57485a8e6176c4dbb68bd1": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HBoxModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HBoxModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HBoxView",
            "box_style": "",
            "children": [
              "IPY_MODEL_7691a29d5bf44d01bafd6643400cbef0",
              "IPY_MODEL_ba580709444348ac99daf82d488239a3",
              "IPY_MODEL_5ddf722463404dbca2e1343722d2fe6b"
            ],
            "layout": "IPY_MODEL_d4491208f1d641adbba7e3d9138aab5a"
          }
        },
        "236c712f6f234e3f8d7f4ea00feb7f6a": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "FloatProgressModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "FloatProgressModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "ProgressView",
            "bar_style": "success",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_e8bd37b87fba4f8a9e2d0e9bb80a13f0",
            "max": 1042301,
            "min": 0,
            "orientation": "horizontal",
            "style": "IPY_MODEL_748e4fce751841348c8907d56a01d2a1",
            "value": 1042301
          }
        },
        "2510e83f724343019d01a35a84e76bf9": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "27fa70cf544147f29b2a8a17c4f2faa3": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HTMLModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_64f492d1096149ad9da667ea864d7f2b",
            "placeholder": "​",
            "style": "IPY_MODEL_bc9c54b508ed4aabad64135937d16a70",
            "value": " 548M/548M [00:07&lt;00:00, 25.0MB/s]"
          }
        },
        "29201ca978c5470aa05877d0ace36f31": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "2923856e65fd40718b7358bfbb3143cf": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "29c0f9d47313413386f284f7393211d1": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "2bb6760048f24cb59f6ff5f16b3c7acd": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "2c3ffd876324474594f95fe1c71d3782": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "2e3a568ae6714c4aab97b67c7db2eb07": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "2e3fa17580e544aaa238a07cdeb116b7": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "FloatProgressModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "FloatProgressModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "ProgressView",
            "bar_style": "success",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_af5827003e7c43d8b6bda1efceeb1fae",
            "max": 1355256,
            "min": 0,
            "orientation": "horizontal",
            "style": "IPY_MODEL_5d82735ec2df45c7b9fd820df07d938d",
            "value": 1355256
          }
        },
        "30ae280e5d5b462daa1f61b929e51a3e": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "329fab4cb7a14053b47bdd56ad38ec60": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HBoxModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HBoxModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HBoxView",
            "box_style": "",
            "children": [
              "IPY_MODEL_d839e15480684df99f1ce3a7c6c41342",
              "IPY_MODEL_774a79f0517a4d5c8b2a037f6a59dc3e",
              "IPY_MODEL_7b52f480da1e4554a39b731269dc5d28"
            ],
            "layout": "IPY_MODEL_96762776b2564b0ca0eda807d998d097"
          }
        },
        "334a2eeec579455ba662deabd578f29d": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "3451ffe86a4a4d7a8858e4e720789857": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "38b91d003fc541908101453b9655abb2": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HTMLModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_d6ba35709a0f4221bff3209b66f6ccde",
            "placeholder": "​",
            "style": "IPY_MODEL_64f0e8ad62e448f9b767515e44bf4e9b",
            "value": " 116722/116722 [00:04&lt;00:00, 17225.77 examples/s]"
          }
        },
        "39e1e4fa159044ea97c5f9c78ba14a1c": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "40465bc3852f40018eb5173d2a789c7f": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "41afb58c5a964028a4943def4b861b97": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "41f16b59ab674b67ab1577adb413ae48": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "44e178a0be524daa95b256d5bff31202": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HTMLModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_452cea6360ce498c85e843e348deb1f4",
            "placeholder": "​",
            "style": "IPY_MODEL_16f10c9d3be140829e228ad4b3aa7705",
            "value": "Downloading data: 100%"
          }
        },
        "452cea6360ce498c85e843e348deb1f4": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "458e2312cf2c47fea736fc230083a0de": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HBoxModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HBoxModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HBoxView",
            "box_style": "",
            "children": [
              "IPY_MODEL_83f9dbdfcf7443ce8589e0c59116afcf",
              "IPY_MODEL_9ed93d180d26497596cdeb3f8d014e09",
              "IPY_MODEL_9544135d58ee491191ea3b16bf2424eb"
            ],
            "layout": "IPY_MODEL_2510e83f724343019d01a35a84e76bf9"
          }
        },
        "45a46219b04b4d5989b5a7a241c165b5": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "4759413c86e845d9a8e374ee135ff414": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "FloatProgressModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "FloatProgressModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "ProgressView",
            "bar_style": "success",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_ab0a3df7a8904389904bfcb31a148b7f",
            "max": 116722,
            "min": 0,
            "orientation": "horizontal",
            "style": "IPY_MODEL_c692217a3ef84e04bccb3bcdd4a2ec8a",
            "value": 116722
          }
        },
        "4929031e60fa4ad883215fc9470113ae": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "4a452ec981ff415680f80c2cfd30a0ad": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "ProgressStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "ProgressStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "bar_color": null,
            "description_width": ""
          }
        },
        "4a9bad66b87348589618e074a6b2f747": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HTMLModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_5cef8de414a447ff911632ff05beedfb",
            "placeholder": "​",
            "style": "IPY_MODEL_39e1e4fa159044ea97c5f9c78ba14a1c",
            "value": "Downloading data: 100%"
          }
        },
        "4b02a2cb6fe14ccd9d85a86994c6cb67": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "5243aeab7e624b02b273463fb32fbf42": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "ProgressStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "ProgressStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "bar_color": null,
            "description_width": ""
          }
        },
        "5446c6060f8e419ab8184d0e15981690": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HTMLModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_a97ec2a948e549ae9d97c2ce6d8d962e",
            "placeholder": "​",
            "style": "IPY_MODEL_b2729e2fde2549d1b531e9320dd6649d",
            "value": "vocab.json: 100%"
          }
        },
        "55634652e74049b199344e325dff5676": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "57be34f94b554f0e94f45635efd721b7": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HTMLModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_7cf3fe343cdd45b291523b15dc8f4c1a",
            "placeholder": "​",
            "style": "IPY_MODEL_ac236fc0d5bc4d8f962e3d35b859893d",
            "value": "tokenizer_config.json: 100%"
          }
        },
        "5835fb83bd014edcaedc1974e2738273": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "58cabc7c155747fc8cbe2f27f1beb289": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "5c38c56a72d74118a65c7eca2d47aaa0": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "5cef8de414a447ff911632ff05beedfb": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "5d1aaf2a41fd47e1ba3a2a73855db66b": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HBoxModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HBoxModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HBoxView",
            "box_style": "",
            "children": [
              "IPY_MODEL_4a9bad66b87348589618e074a6b2f747",
              "IPY_MODEL_94712aabc9504ec7a3d204a914118b6c",
              "IPY_MODEL_70010175a0c9460c89a2297895f9425e"
            ],
            "layout": "IPY_MODEL_9cf51736a40e490ba9daa0aa4585d614"
          }
        },
        "5d82735ec2df45c7b9fd820df07d938d": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "ProgressStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "ProgressStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "bar_color": null,
            "description_width": ""
          }
        },
        "5d88dfa6c6e04cc689bdc9f753fd26fa": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HBoxModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HBoxModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HBoxView",
            "box_style": "",
            "children": [
              "IPY_MODEL_bb4c6b6b74824b6b8fcc74009aa36768",
              "IPY_MODEL_c6acfb70629745729fa440a02b5aa6f1",
              "IPY_MODEL_d4c3590ed00243a99495add140c2f504"
            ],
            "layout": "IPY_MODEL_4b02a2cb6fe14ccd9d85a86994c6cb67"
          }
        },
        "5dd2682ec2ab4a1ba0d0023ffe3cc4a9": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "5ddf722463404dbca2e1343722d2fe6b": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HTMLModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_e47ec19a85da4dbaab999b5bb6490c05",
            "placeholder": "​",
            "style": "IPY_MODEL_2e3a568ae6714c4aab97b67c7db2eb07",
            "value": " 124/124 [00:00&lt;00:00, 7.33kB/s]"
          }
        },
        "5e9e490c3c244b09a1d8f9007a5d8717": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "646ee18fffee4f0c89f23871c56223a8": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "64f0e8ad62e448f9b767515e44bf4e9b": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "64f492d1096149ad9da667ea864d7f2b": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "6e2efc4590694d2ab3e02f4baa2ed821": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "70010175a0c9460c89a2297895f9425e": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HTMLModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_f61f3b244ab5421e8a2c18d37c33e290",
            "placeholder": "​",
            "style": "IPY_MODEL_920ac1c6300d4ac2b826fa0fcc1c4cd8",
            "value": " 111M/111M [00:00&lt;00:00, 171MB/s]"
          }
        },
        "711cce8380d8411181387e51b7da25bb": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "7151d446f6964946af694275001a8b50": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "723a0a0fae6344b2bdc445d69af1517a": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HTMLModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_c16c6f1424134cb7ab354757b7ce2c15",
            "placeholder": "​",
            "style": "IPY_MODEL_0b831cc698eb4838960af0add2becea1",
            "value": " 1.36M/1.36M [00:00&lt;00:00, 37.9MB/s]"
          }
        },
        "748e4fce751841348c8907d56a01d2a1": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "ProgressStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "ProgressStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "bar_color": null,
            "description_width": ""
          }
        },
        "7691a29d5bf44d01bafd6643400cbef0": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HTMLModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_8bbdbacf56ee44aca93e9a9a7ea824e0",
            "placeholder": "​",
            "style": "IPY_MODEL_711cce8380d8411181387e51b7da25bb",
            "value": "generation_config.json: 100%"
          }
        },
        "774a79f0517a4d5c8b2a037f6a59dc3e": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "FloatProgressModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "FloatProgressModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "ProgressView",
            "bar_style": "success",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_4929031e60fa4ad883215fc9470113ae",
            "max": 6553,
            "min": 0,
            "orientation": "horizontal",
            "style": "IPY_MODEL_a6bc156f830b4829b35f298ed7d733bf",
            "value": 6553
          }
        },
        "7b52f480da1e4554a39b731269dc5d28": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HTMLModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_6e2efc4590694d2ab3e02f4baa2ed821",
            "placeholder": "​",
            "style": "IPY_MODEL_de83309c1bd54e66a613b755a89f4455",
            "value": " 6553/6553 [00:00&lt;00:00, 66960.49 examples/s]"
          }
        },
        "7b6d4a6e48be478a8f694bccbdfaa81d": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "7b770a26b25c449fbdac9824c6b00204": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HTMLModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_8cc35a3254b049de9d1535e4c4a64dda",
            "placeholder": "​",
            "style": "IPY_MODEL_5835fb83bd014edcaedc1974e2738273",
            "value": " 1.04M/1.04M [00:00&lt;00:00, 12.0MB/s]"
          }
        },
        "7bcdb14ff8c64ff4b1d7b882a85f89dd": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HTMLModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_005b8f782bdb4ab2a8a5e40a75674d96",
            "placeholder": "​",
            "style": "IPY_MODEL_5c38c56a72d74118a65c7eca2d47aaa0",
            "value": "tokenizer.json: 100%"
          }
        },
        "7cd8e757c96f4a4481c5fffd99fb2cf2": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HBoxModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HBoxModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HBoxView",
            "box_style": "",
            "children": [
              "IPY_MODEL_c5541f7a0b3d405388df008eb820a130",
              "IPY_MODEL_ca480988b2404679954759cb9e2c991c",
              "IPY_MODEL_f73da3dc44384c7baad205eaa14bfb41"
            ],
            "layout": "IPY_MODEL_af155570f61742c181961e49703cf911"
          }
        },
        "7cf3fe343cdd45b291523b15dc8f4c1a": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "7d3ece0ec61a4f7eb21808f759df7690": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "83f9dbdfcf7443ce8589e0c59116afcf": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HTMLModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_c8cc0f0cf5d14e83b002ead2a0c3afa4",
            "placeholder": "​",
            "style": "IPY_MODEL_de3fe2d2e22547e1beb6c76b59d71082",
            "value": "Downloading data: 100%"
          }
        },
        "8525e5dc75cb42d790ee5e59c96cee5f": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HTMLModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_19a4391701ad4399887e35757ed7eea6",
            "placeholder": "​",
            "style": "IPY_MODEL_7b6d4a6e48be478a8f694bccbdfaa81d",
            "value": "model.safetensors: 100%"
          }
        },
        "890cde66805f46b2bcc713df00dfbe22": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "ProgressStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "ProgressStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "bar_color": null,
            "description_width": ""
          }
        },
        "896bed8372804158aa4bc944e6401f42": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HTMLModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_41f16b59ab674b67ab1577adb413ae48",
            "placeholder": "​",
            "style": "IPY_MODEL_1dd9d180974942d89cabcc4d318a6cf0",
            "value": " 665/665 [00:00&lt;00:00, 33.9kB/s]"
          }
        },
        "8bbdbacf56ee44aca93e9a9a7ea824e0": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "8cb3dd49227d4a72b4d7938444f0202a": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "ProgressStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "ProgressStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "bar_color": null,
            "description_width": ""
          }
        },
        "8cc35a3254b049de9d1535e4c4a64dda": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "8d354fed57014964962451731bb319ef": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HTMLModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_93c404b763094295aa1bbcdc5847e3b9",
            "placeholder": "​",
            "style": "IPY_MODEL_2c3ffd876324474594f95fe1c71d3782",
            "value": "Downloading readme: 100%"
          }
        },
        "8ee845415acf4bb29a6a3fc5c2c72070": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "ProgressStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "ProgressStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "bar_color": null,
            "description_width": ""
          }
        },
        "90066549c0b4446bbae7b2a16fbe1a6b": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "920ac1c6300d4ac2b826fa0fcc1c4cd8": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "93c404b763094295aa1bbcdc5847e3b9": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "94712aabc9504ec7a3d204a914118b6c": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "FloatProgressModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "FloatProgressModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "ProgressView",
            "bar_style": "success",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_45a46219b04b4d5989b5a7a241c165b5",
            "max": 110623451,
            "min": 0,
            "orientation": "horizontal",
            "style": "IPY_MODEL_a5c0c29181fc4821b4017e7028372431",
            "value": 110623451
          }
        },
        "9544135d58ee491191ea3b16bf2424eb": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HTMLModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_90066549c0b4446bbae7b2a16fbe1a6b",
            "placeholder": "​",
            "style": "IPY_MODEL_18439c99c16d4162b8a2a52d86dc5847",
            "value": " 6.23M/6.23M [00:00&lt;00:00, 54.1MB/s]"
          }
        },
        "96762776b2564b0ca0eda807d998d097": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "97816036a7c643feb7c24755f4e2233b": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "99d0e034709c40359095a9004cc46798": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "9b3cc7be38ab486abb29f68c4ec58388": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "9cf51736a40e490ba9daa0aa4585d614": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "9ed93d180d26497596cdeb3f8d014e09": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "FloatProgressModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "FloatProgressModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "ProgressView",
            "bar_style": "success",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_d1fe338ae4604838b403b255a4637154",
            "max": 6225522,
            "min": 0,
            "orientation": "horizontal",
            "style": "IPY_MODEL_5243aeab7e624b02b273463fb32fbf42",
            "value": 6225522
          }
        },
        "a22888ab49b941bfae153b1d7d6a3fe9": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HBoxModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HBoxModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HBoxView",
            "box_style": "",
            "children": [
              "IPY_MODEL_054c5f4fc64744a7a46e991855c7b386",
              "IPY_MODEL_caaca6163cff4afb9b28978b1e64065a",
              "IPY_MODEL_896bed8372804158aa4bc944e6401f42"
            ],
            "layout": "IPY_MODEL_40465bc3852f40018eb5173d2a789c7f"
          }
        },
        "a483b3f8216d4de88866cb1a017c14bb": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "a4d54d72871743a68f60983a3bdbf55d": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "a5c0c29181fc4821b4017e7028372431": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "ProgressStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "ProgressStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "bar_color": null,
            "description_width": ""
          }
        },
        "a6bc156f830b4829b35f298ed7d733bf": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "ProgressStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "ProgressStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "bar_color": null,
            "description_width": ""
          }
        },
        "a97ec2a948e549ae9d97c2ce6d8d962e": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "ab0a3df7a8904389904bfcb31a148b7f": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "ac236fc0d5bc4d8f962e3d35b859893d": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "ac756657854e4ec78993df53314cc18f": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HTMLModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_ed16f76131904795900803fb1238f73d",
            "placeholder": "​",
            "style": "IPY_MODEL_334a2eeec579455ba662deabd578f29d",
            "value": " 6.12M/6.12M [00:00&lt;00:00, 60.9MB/s]"
          }
        },
        "af155570f61742c181961e49703cf911": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "af5827003e7c43d8b6bda1efceeb1fae": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "b2729e2fde2549d1b531e9320dd6649d": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "b41bb97a0a7a4cd589d837144eed8200": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HBoxModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HBoxModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HBoxView",
            "box_style": "",
            "children": [
              "IPY_MODEL_5446c6060f8e419ab8184d0e15981690",
              "IPY_MODEL_236c712f6f234e3f8d7f4ea00feb7f6a",
              "IPY_MODEL_7b770a26b25c449fbdac9824c6b00204"
            ],
            "layout": "IPY_MODEL_30ae280e5d5b462daa1f61b929e51a3e"
          }
        },
        "b54c45a49eb1484194a576d0421c0fe0": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HTMLModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_a483b3f8216d4de88866cb1a017c14bb",
            "placeholder": "​",
            "style": "IPY_MODEL_3451ffe86a4a4d7a8858e4e720789857",
            "value": " 532/532 [00:00&lt;00:00, 14.7kB/s]"
          }
        },
        "b6fa7a2add97484cbea0c26ed46d29e6": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HBoxModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HBoxModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HBoxView",
            "box_style": "",
            "children": [
              "IPY_MODEL_44e178a0be524daa95b256d5bff31202",
              "IPY_MODEL_c33b7e89df304f78b3b498a8f9c68383",
              "IPY_MODEL_ac756657854e4ec78993df53314cc18f"
            ],
            "layout": "IPY_MODEL_2923856e65fd40718b7358bfbb3143cf"
          }
        },
        "ba580709444348ac99daf82d488239a3": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "FloatProgressModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "FloatProgressModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "ProgressView",
            "bar_style": "success",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_bf8df483688b4650bb5279a23602a520",
            "max": 124,
            "min": 0,
            "orientation": "horizontal",
            "style": "IPY_MODEL_dbe917ae193a4c3c8cbcc9254dd8f639",
            "value": 124
          }
        },
        "bb4c6b6b74824b6b8fcc74009aa36768": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HTMLModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_a4d54d72871743a68f60983a3bdbf55d",
            "placeholder": "​",
            "style": "IPY_MODEL_0e017c1b69fe4e89b3a5e169c9387445",
            "value": "Generating valid split: 100%"
          }
        },
        "bc9c54b508ed4aabad64135937d16a70": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "bd1e9d756633421e980ac84a1ed3df96": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HTMLModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_99d0e034709c40359095a9004cc46798",
            "placeholder": "​",
            "style": "IPY_MODEL_e6953b8a8ace427ca7689f7fc6cfd6b3",
            "value": " 26.0/26.0 [00:00&lt;00:00, 1.39kB/s]"
          }
        },
        "bda44f920ef0452991f345ffb8843f73": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "bf8df483688b4650bb5279a23602a520": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "c16c6f1424134cb7ab354757b7ce2c15": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "c33b7e89df304f78b3b498a8f9c68383": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "FloatProgressModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "FloatProgressModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "ProgressView",
            "bar_style": "success",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_bda44f920ef0452991f345ffb8843f73",
            "max": 6124527,
            "min": 0,
            "orientation": "horizontal",
            "style": "IPY_MODEL_890cde66805f46b2bcc713df00dfbe22",
            "value": 6124527
          }
        },
        "c5541f7a0b3d405388df008eb820a130": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HTMLModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_9b3cc7be38ab486abb29f68c4ec58388",
            "placeholder": "​",
            "style": "IPY_MODEL_5e9e490c3c244b09a1d8f9007a5d8717",
            "value": "merges.txt: 100%"
          }
        },
        "c692217a3ef84e04bccb3bcdd4a2ec8a": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "ProgressStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "ProgressStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "bar_color": null,
            "description_width": ""
          }
        },
        "c6acfb70629745729fa440a02b5aa6f1": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "FloatProgressModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "FloatProgressModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "ProgressView",
            "bar_style": "success",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_5dd2682ec2ab4a1ba0d0023ffe3cc4a9",
            "max": 6447,
            "min": 0,
            "orientation": "horizontal",
            "style": "IPY_MODEL_4a452ec981ff415680f80c2cfd30a0ad",
            "value": 6447
          }
        },
        "c6c5a944a23e456ebb6c69805e4d4e2e": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "c8cc0f0cf5d14e83b002ead2a0c3afa4": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "ca480988b2404679954759cb9e2c991c": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "FloatProgressModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "FloatProgressModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "ProgressView",
            "bar_style": "success",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_97816036a7c643feb7c24755f4e2233b",
            "max": 456318,
            "min": 0,
            "orientation": "horizontal",
            "style": "IPY_MODEL_8ee845415acf4bb29a6a3fc5c2c72070",
            "value": 456318
          }
        },
        "caa2f6b29a0349aa80d773185181056b": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HBoxModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HBoxModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HBoxView",
            "box_style": "",
            "children": [
              "IPY_MODEL_8d354fed57014964962451731bb319ef",
              "IPY_MODEL_0faf1d3472974ceca6eeda7e9d5f7906",
              "IPY_MODEL_b54c45a49eb1484194a576d0421c0fe0"
            ],
            "layout": "IPY_MODEL_2bb6760048f24cb59f6ff5f16b3c7acd"
          }
        },
        "caaca6163cff4afb9b28978b1e64065a": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "FloatProgressModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "FloatProgressModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "ProgressView",
            "bar_style": "success",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_12ce1164d18249b3807e5ec45aac4a84",
            "max": 665,
            "min": 0,
            "orientation": "horizontal",
            "style": "IPY_MODEL_1f99bdeb99dc43f882494cb9a8d1f664",
            "value": 665
          }
        },
        "cce757f48382449da0d3e3bf87272284": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "ProgressStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "ProgressStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "bar_color": null,
            "description_width": ""
          }
        },
        "cf4a4ec0a7124457a68d70970889b744": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "d18ab3331f1a42f1af89aca50b9e78e6": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "d1fe338ae4604838b403b255a4637154": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "d3fccbdf94bc44cdb552e3ac37a54e0f": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HBoxModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HBoxModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HBoxView",
            "box_style": "",
            "children": [
              "IPY_MODEL_ddcd794463174f51b09f6b77ab732b94",
              "IPY_MODEL_4759413c86e845d9a8e374ee135ff414",
              "IPY_MODEL_38b91d003fc541908101453b9655abb2"
            ],
            "layout": "IPY_MODEL_7d3ece0ec61a4f7eb21808f759df7690"
          }
        },
        "d4491208f1d641adbba7e3d9138aab5a": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "d4c3590ed00243a99495add140c2f504": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HTMLModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_e53325f28cab44a5ab6a9218764b45d2",
            "placeholder": "​",
            "style": "IPY_MODEL_d5f193695fde47e8b9edd4f84f2ecf2b",
            "value": " 6447/6447 [00:00&lt;00:00, 74605.88 examples/s]"
          }
        },
        "d5f193695fde47e8b9edd4f84f2ecf2b": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "d6ba35709a0f4221bff3209b66f6ccde": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "d7554a7bb8bd4d5dbbc03268dc2d77c4": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "d839e15480684df99f1ce3a7c6c41342": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HTMLModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_d7554a7bb8bd4d5dbbc03268dc2d77c4",
            "placeholder": "​",
            "style": "IPY_MODEL_cf4a4ec0a7124457a68d70970889b744",
            "value": "Generating test split: 100%"
          }
        },
        "dbe917ae193a4c3c8cbcc9254dd8f639": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "ProgressStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "ProgressStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "bar_color": null,
            "description_width": ""
          }
        },
        "ddcd794463174f51b09f6b77ab732b94": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HTMLModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_29201ca978c5470aa05877d0ace36f31",
            "placeholder": "​",
            "style": "IPY_MODEL_7151d446f6964946af694275001a8b50",
            "value": "Generating train split: 100%"
          }
        },
        "de3fe2d2e22547e1beb6c76b59d71082": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "de83309c1bd54e66a613b755a89f4455": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "dfc95e5b17b94e1dae5b0c5a26a9da5d": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HBoxModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HBoxModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HBoxView",
            "box_style": "",
            "children": [
              "IPY_MODEL_7bcdb14ff8c64ff4b1d7b882a85f89dd",
              "IPY_MODEL_2e3fa17580e544aaa238a07cdeb116b7",
              "IPY_MODEL_723a0a0fae6344b2bdc445d69af1517a"
            ],
            "layout": "IPY_MODEL_29c0f9d47313413386f284f7393211d1"
          }
        },
        "e400bbe893d744b4bc4ec3c00d0ab49a": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HBoxModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HBoxModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HBoxView",
            "box_style": "",
            "children": [
              "IPY_MODEL_57be34f94b554f0e94f45635efd721b7",
              "IPY_MODEL_04588e43dc174ae5bcebc0ef01d581dc",
              "IPY_MODEL_bd1e9d756633421e980ac84a1ed3df96"
            ],
            "layout": "IPY_MODEL_ed839cf80d8e4f4a8feaa109cd9202b6"
          }
        },
        "e47ec19a85da4dbaab999b5bb6490c05": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "e53325f28cab44a5ab6a9218764b45d2": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "e6953b8a8ace427ca7689f7fc6cfd6b3": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "DescriptionStyleModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "StyleView",
            "description_width": ""
          }
        },
        "e8bd37b87fba4f8a9e2d0e9bb80a13f0": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "ed16f76131904795900803fb1238f73d": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "ed839cf80d8e4f4a8feaa109cd9202b6": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "f10cd68089cc4e3ebe9b774054517c80": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "FloatProgressModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "FloatProgressModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "ProgressView",
            "bar_style": "success",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_646ee18fffee4f0c89f23871c56223a8",
            "max": 548105171,
            "min": 0,
            "orientation": "horizontal",
            "style": "IPY_MODEL_8cb3dd49227d4a72b4d7938444f0202a",
            "value": 548105171
          }
        },
        "f61f3b244ab5421e8a2c18d37c33e290": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "f73da3dc44384c7baad205eaa14bfb41": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HTMLModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HTMLModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HTMLView",
            "description": "",
            "description_tooltip": null,
            "layout": "IPY_MODEL_41afb58c5a964028a4943def4b861b97",
            "placeholder": "​",
            "style": "IPY_MODEL_55634652e74049b199344e325dff5676",
            "value": " 456k/456k [00:00&lt;00:00, 21.4MB/s]"
          }
        },
        "fe88b66639f64622adda4094cb268fa3": {
          "model_module": "@jupyter-widgets/base",
          "model_module_version": "1.2.0",
          "model_name": "LayoutModel",
          "state": {
            "_model_module": "@jupyter-widgets/base",
            "_model_module_version": "1.2.0",
            "_model_name": "LayoutModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/base",
            "_view_module_version": "1.2.0",
            "_view_name": "LayoutView",
            "align_content": null,
            "align_items": null,
            "align_self": null,
            "border": null,
            "bottom": null,
            "display": null,
            "flex": null,
            "flex_flow": null,
            "grid_area": null,
            "grid_auto_columns": null,
            "grid_auto_flow": null,
            "grid_auto_rows": null,
            "grid_column": null,
            "grid_gap": null,
            "grid_row": null,
            "grid_template_areas": null,
            "grid_template_columns": null,
            "grid_template_rows": null,
            "height": null,
            "justify_content": null,
            "justify_items": null,
            "left": null,
            "margin": null,
            "max_height": null,
            "max_width": null,
            "min_height": null,
            "min_width": null,
            "object_fit": null,
            "object_position": null,
            "order": null,
            "overflow": null,
            "overflow_x": null,
            "overflow_y": null,
            "padding": null,
            "right": null,
            "top": null,
            "visibility": null,
            "width": null
          }
        },
        "ff6e07c2c9de408cb353b2b57ad52673": {
          "model_module": "@jupyter-widgets/controls",
          "model_module_version": "1.5.0",
          "model_name": "HBoxModel",
          "state": {
            "_dom_classes": [],
            "_model_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_model_name": "HBoxModel",
            "_view_count": null,
            "_view_module": "@jupyter-widgets/controls",
            "_view_module_version": "1.5.0",
            "_view_name": "HBoxView",
            "box_style": "",
            "children": [
              "IPY_MODEL_8525e5dc75cb42d790ee5e59c96cee5f",
              "IPY_MODEL_f10cd68089cc4e3ebe9b774054517c80",
              "IPY_MODEL_27fa70cf544147f29b2a8a17c4f2faa3"
            ],
            "layout": "IPY_MODEL_fe88b66639f64622adda4094cb268fa3"
          }
        }
      }
    }
  },
  "nbformat": 4,
  "nbformat_minor": 0
}