Story_Generation_DistilGPT.ipynb

{
  "cells": [
    {
      "cell_type": "markdown",
      "id": "02538164",
      "metadata": {
        "id": "02538164"
      },
      "source": [
        "## Importing Necessary Libraries"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "ccf66873",
      "metadata": {
        "id": "ccf66873"
      },
      "outputs": [],
      "source": [
        "import pandas as pd\n",
        "import numpy as np\n",
        "import random\n",
        "import torch\n",
        "from transformers import AutoTokenizer, DataCollatorForLanguageModeling, AutoModelForCausalLM, TrainingArguments, Trainer\n",
        "from tqdm import tqdm, trange\n",
        "import math"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "a41acde8",
      "metadata": {
        "id": "a41acde8"
      },
      "source": [
        "## Model and Tokenizer Initialization"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "2e4a8d27",
      "metadata": {
        "id": "2e4a8d27"
      },
      "outputs": [],
      "source": [
        "base_model = AutoModelForCausalLM.from_pretrained(\"distilgpt2\")\n",
        "base_tokenizer = AutoTokenizer.from_pretrained('distilgpt2')"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "530a46f3",
      "metadata": {
        "id": "530a46f3"
      },
      "source": [
        "## Data Preprocessing"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "122119d3",
      "metadata": {
        "id": "122119d3"
      },
      "outputs": [],
      "source": [
        "def combine(x, base_tokenizer=base_tokenizer):\n",
        "    return x[\"Input\"]+\" \"+\"Summary: \"+x['Summary']"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "1403062f",
      "metadata": {
        "id": "1403062f"
      },
      "outputs": [],
      "source": [
        "import pandas as pd\n",
        "df = pd.read_csv('/content/drive/MyDrive/Visual Story Telling/Dataset - Story Generation/Plot_Summary_Dataset')"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "4e3d026d",
      "metadata": {
        "id": "4e3d026d"
      },
      "outputs": [],
      "source": [
        "df['Input'] = df.apply(combine, axis=1)\n",
        "df = df['Input']\n",
        "df.to_csv('/content/drive/MyDrive/Visual Story Telling/Dataset - Story Generation/Plot_Summary_DistilGPT2', index=False)"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "## Data Split"
      ],
      "metadata": {
        "id": "RpTA0yPprG7R"
      },
      "id": "RpTA0yPprG7R"
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "4c90d107",
      "metadata": {
        "id": "4c90d107"
      },
      "outputs": [],
      "source": [
        "def encode_sequences(x, base_tokenizer = base_tokenizer):\n",
        "    # Max length of the input sequence in DistilGPT2 is 1024 tokens\n",
        "    return base_tokenizer(x, max_length = 1024, truncation=True, add_special_tokens = True)['input_ids']"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "import pandas as pd\n",
        "df = pd.read_csv('/content/drive/MyDrive/Visual Story Telling/Dataset - Story Generation/Plot_Summary_DistilGPT2')"
      ],
      "metadata": {
        "id": "QmUO0NlJrMSA"
      },
      "id": "QmUO0NlJrMSA",
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "5a8632c4",
      "metadata": {
        "id": "5a8632c4"
      },
      "outputs": [],
      "source": [
        "from sklearn.model_selection import train_test_split\n",
        "train_df, test_df = train_test_split(df, test_size=0.1, random_state=1)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "20cb639d",
      "metadata": {
        "id": "20cb639d"
      },
      "outputs": [],
      "source": [
        "train_df = train_df['Input'].apply(encode_sequences)\n",
        "test_df = test_df['Input'].apply(encode_sequences)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "23d49a95",
      "metadata": {
        "id": "23d49a95"
      },
      "outputs": [],
      "source": [
        "train_df.dropna(inplace = True)\n",
        "test_df.dropna(inplace = True)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "5354f0fd",
      "metadata": {
        "id": "5354f0fd"
      },
      "outputs": [],
      "source": [
        "train_df.reset_index(drop=True, inplace=True)\n",
        "test_df.reset_index(drop=True, inplace=True)"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "f0dce2a5",
      "metadata": {
        "id": "f0dce2a5"
      },
      "source": [
        "## Model Training"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "b562a478",
      "metadata": {
        "id": "b562a478"
      },
      "outputs": [],
      "source": [
        "#Setting pad token same as eos token\n",
        "base_tokenizer.pad_token = base_tokenizer.eos_token\n",
        "\n",
        "#Initializing Data Collator that forms batches and sends the input in a proper language modeling format to the model for training and evaluation\n",
        "data_collator = DataCollatorForLanguageModeling(\n",
        "        tokenizer=base_tokenizer,\n",
        "        mlm=False,\n",
        "        return_tensors=\"pt\"\n",
        "    )"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "c8e2ed6c",
      "metadata": {
        "id": "c8e2ed6c"
      },
      "outputs": [],
      "source": [
        "model_path = '/content/drive/MyDrive/Visual Story Telling/Story_Gen_Model/DistilGPT2'\n",
        "\n",
        "#Specifying the Training arguments for the model training\n",
        "training_args = TrainingArguments(\n",
        "    output_dir=model_path,\n",
        "    overwrite_output_dir = True,\n",
        "    evaluation_strategy=\"no\",\n",
        "    gradient_accumulation_steps=8,\n",
        "    num_train_epochs=30,\n",
        "    weight_decay=0.01,\n",
        "    lr_scheduler_type=\"cosine\",\n",
        "    learning_rate=5e-4,\n",
        "    fp16=True\n",
        ")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "0fa2f49b",
      "metadata": {
        "id": "0fa2f49b"
      },
      "outputs": [],
      "source": [
        "# Initializing the trainer\n",
        "\n",
        "trainer = Seq2SeqTrainer(\n",
        "    model=base_model,                         # the instantiated  Transformers model to be trained\n",
        "    args=training_args,                  # training arguments, defined above\n",
        "    data_collator=data_collator,\n",
        "    train_dataset=train_df,       # training dataset\n",
        "    eval_dataset = test_df,\n",
        ")"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "import gc\n",
        "gc.collect()\n",
        "torch.cuda.empty_cache()"
      ],
      "metadata": {
        "id": "GC2Xh931OcD3"
      },
      "id": "GC2Xh931OcD3",
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "3fb1918c",
      "metadata": {
        "id": "3fb1918c"
      },
      "outputs": [],
      "source": [
        "# Starting the training\n",
        "trainer.train()"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "20fd75ae",
      "metadata": {
        "id": "20fd75ae"
      },
      "outputs": [],
      "source": [
        "# Saving the final model\n",
        "trainer.save_model()"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "f4880a22",
      "metadata": {
        "id": "f4880a22"
      },
      "source": [
        "## Model Evaluation (Perplexity)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "e093d1d8",
      "metadata": {
        "id": "e093d1d8"
      },
      "outputs": [],
      "source": [
        "eval_results = trainer.evaluate()"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "print(f\"Perplexity: {math.exp(eval_results['eval_loss']):.2f}\")"
      ],
      "metadata": {
        "id": "xhbatHObrtPq"
      },
      "id": "xhbatHObrtPq",
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "id": "9e098d0e",
      "metadata": {
        "id": "9e098d0e"
      },
      "source": [
        "## Loading and Testing"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "10acf61d",
      "metadata": {
        "id": "10acf61d"
      },
      "outputs": [],
      "source": [
        "base_model = AutoModelForCausalLM.from_pretrained(\"/content/drive/MyDrive/Visual Story Telling/Story_Gen_Model/DistilGPT2\")\n",
        "base_tokenizer = AutoTokenizer.from_pretrained('distilgpt2')"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "b44662c0",
      "metadata": {
        "id": "b44662c0"
      },
      "outputs": [],
      "source": [
        "baseline_tokenizer.pad_token = baseline_tokenizer.eos_token"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "92cd96eb",
      "metadata": {
        "id": "92cd96eb"
      },
      "outputs": [],
      "source": [
        "\"\"\"\n",
        "Format of the input\n",
        "Title: ADD_TITLE. Genre: ADD_GENRES. Characters: ADD_CHARS. Relations: Neutral: <>. Positive: <>. Negative: <>.  Plot: ADD_PLOT\n",
        "\"\"\"\n",
        "\n",
        "from transformers import pipeline\n",
        "\n",
        "input = \"Title: Saiyan Warrior. Genre: Action, Adventure, Drama. Characters: Goku, Vegeta, Beerus. Relations: Neutral: <Beerus, Vegeta>. Positive: <Goku, Vegeta>. Negative: <Goku, Beerus>. Plot: Beerus is a god of destruction, who came to Earth to destroy it. Goku and Vegeta in order to protect their family fight and defeat him.\"\n",
        "generator = pipeline(\"text-generation\", model=base_model, tokenizer = base_tokenizer)\n",
        "do_sample = False\n",
        "num_beams = 1\n",
        "top_p = 0.9\n",
        "generator(input, num_beams = num_beams, top_p = top_p, do_sample = do_sample, min_length = 150, max_length = 250)[0]['generated_text'][len(input)+1:]"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "64fcea26",
      "metadata": {
        "id": "64fcea26"
      },
      "source": [
        "## BLEU Score Evaluation"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "49f6c44a",
      "metadata": {
        "id": "49f6c44a"
      },
      "outputs": [],
      "source": [
        "import evaluate\n",
        "bleu = evaluate.load('bleu')"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "base_model = AutoModelForCausalLM.from_pretrained(\"/content/drive/MyDrive/Visual Story Telling/Story_Gen_Model/DistilGPT2\")\n",
        "base_tokenizer = AutoTokenizer.from_pretrained('distilgpt2')\n",
        "base_tokenizer.pad_token = base_tokenizer.eos_token"
      ],
      "metadata": {
        "id": "O166sGENOwLp"
      },
      "id": "O166sGENOwLp",
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "def encode_input(x, base_tokenizer = base_tokenizer):\n",
        "  x = str(x)\n",
        "  idx = x.index(\"Summary:\")\n",
        "  input = x[:idx]\n",
        "  summary = x[idx+9:]\n",
        "  return (input, summary, len(summary))"
      ],
      "metadata": {
        "id": "RwL_BNQeO1gZ"
      },
      "id": "RwL_BNQeO1gZ",
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "df = pd.read_csv('/content/drive/MyDrive/Visual Story Telling/Dataset - Story Generation/Plot_Summary_DistilGPT2')"
      ],
      "metadata": {
        "id": "EmMIu7pbO2pv"
      },
      "id": "EmMIu7pbO2pv",
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "from sklearn.model_selection import train_test_split\n",
        "df.drop(['Unnamed: 0'], axis=1, inplace=True)\n",
        "train_df, test_df = train_test_split(df, test_size=0.1, random_state=1)\n",
        "test_df.reset_index(drop=True, inplace=True)"
      ],
      "metadata": {
        "id": "buz6KvAPO6Vm"
      },
      "id": "buz6KvAPO6Vm",
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "test_df = test_df.assign(Summary=\"\", Len=0)"
      ],
      "metadata": {
        "id": "jBkRIAmJO7a_"
      },
      "id": "jBkRIAmJO7a_",
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "for idx, row in test_df.iterrows():\n",
        "    test_df['Input'].iloc[idx], test_df['Summary'].iloc[idx], test_df['Len'].iloc[idx] = encode_input(test_df['Input'].iloc[idx])"
      ],
      "metadata": {
        "id": "j9W89v28O8l5"
      },
      "id": "j9W89v28O8l5",
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "test_df = test_df[test_df['Len']<350]"
      ],
      "metadata": {
        "id": "DslHtvkeO-GX"
      },
      "id": "DslHtvkeO-GX",
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "test_df.reset_index(drop=True, inplace=True)"
      ],
      "metadata": {
        "id": "jP9Unw1fO_kj"
      },
      "id": "jP9Unw1fO_kj",
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "#Greedy, Ancestral and Nucleus\n",
        "decoding_strats = [[1,1.0, False],[1,1.0, True],[1,0.9, True]]"
      ],
      "metadata": {
        "id": "pfKCnGw5PBDS"
      },
      "id": "pfKCnGw5PBDS",
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "import warnings\n",
        "warnings.simplefilter(\"ignore\")\n",
        "from transformers import pipeline\n",
        "generator = pipeline(\"text-generation\", model=base_model, tokenizer = base_tokenizer)\n",
        "results = []\n",
        "for x in decoding_strats:\n",
        "  predictions = []\n",
        "  references = []\n",
        "  num_beams, top_p, do_sample = x\n",
        "  for idx,row in test_df.iterrows():\n",
        "    references.append(test_df['Summary'].iloc[idx])\n",
        "    predictions.append(generator(test_df['Input'].iloc[idx], num_beams = num_beams, top_p = top_p, do_sample = do_sample, max_length = len(test_df['Input'].iloc[idx])+test_df[\"Len\"].iloc[idx])[0]['generated_text'][len(test_df['Input'].iloc[idx])+10:])\n",
        "  base_results = bleu.compute(predictions=predictions, references=references)\n",
        "  results.append(base_results)\n",
        "print(results)"
      ],
      "metadata": {
        "id": "6Sz5JGuyPCbP"
      },
      "id": "6Sz5JGuyPCbP",
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "id": "14b3fe47",
      "metadata": {
        "id": "14b3fe47"
      },
      "source": [
        "## Loss Curve"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 2,
      "id": "dad7b725",
      "metadata": {
        "id": "dad7b725"
      },
      "outputs": [],
      "source": [
        "import matplotlib.pyplot as plt\n",
        "%matplotlib inline"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 5,
      "id": "8115afa2",
      "metadata": {
        "id": "8115afa2"
      },
      "outputs": [],
      "source": [
        "df = pd.read_json('/content/drive/MyDrive/Visual Story Telling/Story_Gen_Model/DistilGPT2/checkpoint-16000/trainer_state.json')"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 6,
      "id": "6c696df1",
      "metadata": {
        "id": "6c696df1"
      },
      "outputs": [],
      "source": [
        "def get_epoch(x):\n",
        "  return x['epoch']\n",
        "\n",
        "def get_loss(x):\n",
        "  return x['loss']"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 7,
      "id": "991e7d31",
      "metadata": {
        "id": "991e7d31"
      },
      "outputs": [],
      "source": [
        "df['epoch_number'] = df['log_history'].apply(get_epoch)\n",
        "df ['loss'] = df['log_history'].apply(get_loss)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 8,
      "id": "2eef7468",
      "metadata": {
        "id": "2eef7468",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 489
        },
        "outputId": "8c236480-90aa-40f4-cd1b-337d55a887ad"
      },
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "Text(0.5, 1.0, 'Training curve DistilGPT2')"
            ]
          },
          "metadata": {},
          "execution_count": 8
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 640x480 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ],
      "source": [
        "plt.plot(df['epoch_number'].values, df['loss'].values)\n",
        "plt.xlabel('#Epochs')\n",
        "plt.ylabel('Loss value')\n",
        "plt.title('Training curve DistilGPT2')"
      ]
    }
  ],
  "metadata": {
    "kernelspec": {
      "display_name": "Python (torch-gpu)",
      "language": "python",
      "name": "torch-gpu"
    },
    "language_info": {
      "codemirror_mode": {
        "name": "ipython",
        "version": 3
      },
      "file_extension": ".py",
      "mimetype": "text/x-python",
      "name": "python",
      "nbconvert_exporter": "python",
      "pygments_lexer": "ipython3",
      "version": "3.9.16"
    },
    "colab": {
      "provenance": []
    }
  },
  "nbformat": 4,
  "nbformat_minor": 5
}