ai-generated-content.ipynb

ai-generated-content.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "ai-generated-content.ipynb",
      "provenance": [],
      "authorship_tag": "ABX9TyM8YEp0mpwiuT92OlguqohJ",
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/gist/svpino/c4d1978f9eac47c59e61c34125c8fad9/ai-generated-content.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "HBMCqUqiDMpE",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 102
        },
        "outputId": "81b0aea6-fcdb-4c35-a2a2-0bc8eee57d5f"
      },
      "source": [
        "# This will install the Transformes repository directly\n",
        "# from Github.\n",
        "!pip install -q git+https://github.com/huggingface/transformers.git"
      ],
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\u001b[K     |████████████████████████████████| 3.0MB 2.8MB/s \n",
            "\u001b[K     |████████████████████████████████| 1.1MB 26.0MB/s \n",
            "\u001b[K     |████████████████████████████████| 890kB 38.9MB/s \n",
            "\u001b[?25h  Building wheel for transformers (setup.py) ... \u001b[?25l\u001b[?25hdone\n",
            "  Building wheel for sacremoses (setup.py) ... \u001b[?25l\u001b[?25hdone\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "PUvscABODQ3q",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import tensorflow as tf\n",
        "import textwrap\n",
        "from transformers import TFGPT2LMHeadModel, GPT2Tokenizer"
      ],
      "execution_count": 12,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "BaYoChPhDhJT",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 85
        },
        "outputId": "61fff41e-8009-4bed-9e7d-a90efb859a9b"
      },
      "source": [
        "# We are going to be using GPT-2 Medium. The first time we run this\n",
        "# cell, the models will be downloaded. They are large.\n",
        "tokenizer = GPT2Tokenizer.from_pretrained(\"gpt2-medium\")\n",
        "model = TFGPT2LMHeadModel.from_pretrained(\"gpt2-medium\")"
      ],
      "execution_count": 14,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "All model checkpoint weights were used when initializing TFGPT2LMHeadModel.\n",
            "\n",
            "All the weights of TFGPT2LMHeadModel were initialized from the model checkpoint at gpt2-medium.\n",
            "If your task is similar to the task the model of the checkpoint was trained on, you can already use TFGPT2LMHeadModel for predictions without further training.\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8az6BrxLDTSC",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# This is the seed text. You can to put here some text that will \n",
        "# help the AI focus on a specific topic. The AI will generate text\n",
        "# that's directly related to this seed.\n",
        "SEED_TEXT = \"\"\"\n",
        "    Clean Code should be as enjoyable as reading a good novel. Clean Code tells a story.\n",
        "    Clean Code is a pleasure to read. It's orderly and elegant. Everything is there for a reason. \n",
        "\"\"\"\n",
        "\n",
        "# This represent the number of words we want to generate.\n",
        "MAX_LENGTH = 250"
      ],
      "execution_count": 27,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "bki6iMqjDsc7",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "056b4187-2d79-4d62-f5f9-367967643cbf"
      },
      "source": [
        "# Here we encode the seed text and generate the text.\n",
        "encoded_text = tokenizer.encode(SEED_TEXT, return_tensors='tf')\n",
        "tf.random.set_seed(1)\n",
        "\n",
        "output = model.generate(\n",
        "    encoded_text,\n",
        "    do_sample=True, \n",
        "    max_length=MAX_LENGTH, \n",
        "    top_k=50, \n",
        "    top_p=0.95, \n",
        "    num_return_sequences=1\n",
        ")"
      ],
      "execution_count": 28,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Setting `pad_token_id` to 50256 (first `eos_token_id`) to generate sequence\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "mb07t44BEMIm",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 153
        },
        "outputId": "9d6c8be3-fece-4a3a-99a0-1cf056cfd52e"
      },
      "source": [
        "# Printing out the generated text.\n",
        "for sample in output:\n",
        "    text = tokenizer.decode(sample, skip_special_tokens=True) \n",
        "    text = text[len(SEED_TEXT):].strip()\n",
        "    print(textwrap.fill(text, 80))"
      ],
      "execution_count": 29,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Clean Code takes some ideas that you have from other languages and adds some\n",
            "extra stuff for a more functional look. You'll also notice some interesting\n",
            "features such as object-oriented design and an introduction to functional\n",
            "programming. And of course, you can look at how the author was able to use these\n",
            "features. I would like to see such a book printed around the world. I'd have\n",
            "loved to have seen the book from the perspective of a programmer, but to read in\n",
            "clean and simple style? That would have been quite nice. The editor might have\n",
            "just been wrong, right? Anyway, thank you for making this review!\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "EGmiZOHhFRrv",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    }
  ]
}