{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "finetuning.ipynb",
      "provenance": [],
      "collapsed_sections": []
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "duPhpC7UYvOb",
        "colab_type": "text"
      },
      "source": [
        "#**Setup**\n",
        "\n",
        "---\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "08KZUbQnhKwE",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "\n",
        "!git clone https://github.com/ai4bharat/indic-bert\n",
        "%cd indic-bert\n",
        "!pip3 install -r requirements.txt\n",
        "%cd ..\n",
        "!mkdir indic-glue outputs"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "u4TbQgpAYrSL",
        "colab_type": "text"
      },
      "source": [
        "#**Download Datasets**\n",
        "---\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "B8Te43TtV9OV",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "\n",
        "% cd indic-glue\n",
        "# Download the dataset -- insert link obtained from https://indicnlp.ai4bharat.org/indic-glue/#downloads\n",
        "!wget https://storage.googleapis.com/ai4bharat-public-indic-nlp-corpora/evaluations/wiki-cloze.tar.gz\n",
        "!tar -xaf wiki-cloze.tar.gz\n",
        "% cd ..\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "cAs6r-QSUosR",
        "colab_type": "text"
      },
      "source": [
        "#**Fine-tune the Model**\n",
        "---\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "vUCIjREpQFhv",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "\n",
        "%cd indic-bert\n",
        "\n",
        "import os\n",
        "\n",
        "from fine_tune.cli import main as finetune_main\n",
        "\n",
        "argvec = ['--lang', 'gu',\n",
        "          '--dataset', 'wiki-cloze', # use the right dataset key, check https://github.com/AI4Bharat/indic-bert/blob/master/fine_tune/cli.py#L10\n",
        "          '--model', 'ai4bharat/indic-bert',\n",
        "          '--iglue_dir', '../indic-glue',\n",
        "          '--output_dir', '../outputs',\n",
        "          '--max_seq_length', '128',\n",
        "          '--learning_rate', '2e-5',\n",
        "          '--num_train_epochs', '3',\n",
        "          '--train_batch_size', '32'\n",
        "]\n",
        "\n",
        "finetune_main(argvec)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "zUh3Vw7SUwMW",
        "colab_type": "text"
      },
      "source": [
        "#**Check the Results**\n",
        "---"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Ic0Qpfl-U0Xw",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "!cat /content/outputs/wiki-cloze/gu-gu/model-ai4bharat-indic-bert/test_results.txt"
      ],
      "execution_count": null,
      "outputs": []
    }
  ]
}