{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "# MultiModalHackVAE Demo\n\n",
        "This notebook demonstrates how to use the MultiModalHackVAE model from CatkinChen/nethack-vae.\n\n",
        "## Installation\n\n",
        "```bash\n",
        "pip install torch transformers huggingface_hub\n",
        "# For NetHack environment (optional):\n",
        "pip install nle\n",
        "```"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {},
      "source": [
        "import torch\n",
        "import numpy as np\n",
        "from huggingface_hub import hf_hub_download\n",
        "import json\n",
        "\n",
        "# Load model config\n",
        "config_path = hf_hub_download(repo_id='CatkinChen/nethack-vae', filename='config.json')\n",
        "with open(config_path, 'r') as f:\n",
        "    config = json.load(f)\n",
        "\n",
        "print('Model Configuration:')\n",
        "for key, value in config.items():\n",
        "    print(f'  {key}: {value}')"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {},
      "source": [
        "# Load the model (you'll need to import your model class)\n",
        "# from your_package import MultiModalHackVAE\n",
        "# model = load_model_from_huggingface('{repo_name}')\n",
        "\n",
        "# Example synthetic data\n",
        "batch_size = 1\n",
        "game_chars = torch.randint(32, 127, (batch_size, 21, 79))\n",
        "game_colors = torch.randint(0, 16, (batch_size, 21, 79))\n",
        "blstats = torch.randn(batch_size, 27)\n",
        "msg_tokens = torch.randint(0, 128, (batch_size, 256))\n",
        "hero_info = torch.randint(0, 10, (batch_size, 4))\n",
        "\n",
        "print('Synthetic data shapes:')\n",
        "print(f'  game_chars: {game_chars.shape}')\n",
        "print(f'  game_colors: {game_colors.shape}')\n",
        "print(f'  blstats: {blstats.shape}')\n",
        "print(f'  msg_tokens: {msg_tokens.shape}')\n",
        "print(f'  hero_info: {hero_info.shape}')"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {},
      "source": [
        "# Encode to latent space\n",
        "# with torch.no_grad():\n",
        "#     output = model(\n",
        "#         glyph_chars=game_chars,\n",
        "#         glyph_colors=game_colors,\n",
        "#         blstats=blstats,\n",
        "#         msg_tokens=msg_tokens,\n",
        "#         hero_info=hero_info\n",
        "#     )\n",
        "#     \n",
        "#     latent_mean = output['mu']\n",
        "#     latent_logvar = output['logvar']\n",
        "#     lowrank_factors = output['lowrank_factors']\n",
        "#     \n",
        "#     print(f'Latent representation shape: {latent_mean.shape}')\n",
        "#     print(f'Latent mean: {latent_mean[0][:5].tolist()}')\n",
        "\n",
        "print('Model inference example (uncomment when model is available)')"
      ]
    }
  ],
  "metadata": {
    "kernelspec": {
      "display_name": "Python 3",
      "language": "python",
      "name": "python3"
    },
    "language_info": {
      "name": "python",
      "version": "3.8+"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 4
}