{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "intro"
      },
      "source": [
        "# EcoGrid-OpenEnv: Train 7B Model with GRPO\n",
        "This notebook allows you to train a powerful 7B parameter model (`Qwen2.5-7B-Instruct`) using **Google Colab's Free T4 GPU**.\n",
        "\n",
        "It uses **Unsloth** for blazing-fast 4-bit quantization and **TRL** for Group Relative Policy Optimization (GRPO).\n",
        "\n",
        "### Instructions:\n",
        "1. Go to `Runtime > Change runtime type` and select **T4 GPU**.\n",
        "2. Click **Run All**.\n",
        "3. Once finished, a `lora_adapter.zip` file will automatically download to your computer.\n",
        "4. Extract it into your project folder so the `app.py` can load it!"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "setup"
      },
      "outputs": [],
      "source": [
        "# 1. Install Unsloth and Dependencies\n",
        "!pip install \"unsloth[colab-new] @ git+https://github.com/unslothai/unsloth.git\"\n",
        "!pip install --no-deps xformers trl peft accelerate bitsandbytes\n",
        "\n",
        "# 2. Clone your GitHub Repository\n",
        "import os\n",
        "# Note: Using your public repository.\n",
        "!git clone https://github.com/DivyankLosse/EcoGrid.git\n",
        "os.chdir('EcoGrid')\n",
        "\n",
        "# 3. Run the Training Script!\n",
        "# We use Qwen2.5-7B-Instruct. We'll train for 1 epoch with 500 samples to keep it under 30 minutes.\n",
        "!python train_unsloth.py --model unsloth/Qwen2.5-7B-Instruct --task hard --epochs 1 --samples 500\n",
        "\n",
        "# 4. Zip the results and download\n",
        "import shutil\n",
        "from google.colab import files\n",
        "\n",
        "print(\"Zipping lora_adapter...\")\n",
        "shutil.make_archive('/content/lora_adapter', 'zip', 'lora_adapter')\n",
        "print(\"Zipping logs...\")\n",
        "shutil.make_archive('/content/logs', 'zip', 'logs')\n",
        "\n",
        "print(\"Downloading files...\")\n",
        "files.download('/content/lora_adapter.zip')\n",
        "files.download('/content/logs.zip')"
      ]
    }
  ],
  "metadata": {
    "accelerator": "GPU",
    "colab": {
      "provenance": []
    },
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    },
    "language_info": {
      "name": "python"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 0
}