{
 "nbformat": 4,
 "nbformat_minor": 5,
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "name": "python",
   "version": "3.11.0"
  }
 },
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/duoan/TorchCode/blob/master/templates/13_gpt2_block.ipynb)\n",
    "\n",
    "# 🔴 Hard: GPT-2 Transformer Block\n",
    "\n",
    "Implement a full **GPT-2 style Transformer block** — combining everything you've learned.\n",
    "\n",
    "### Architecture (Pre-Norm)\n",
    "```\n",
    "x = x + causal_self_attention(ln1(x))\n",
    "x = x + mlp(ln2(x))\n",
    "```\n",
    "\n",
    "### Signature\n",
    "```python\n",
    "class GPT2Block(nn.Module):\n",
    "    def __init__(self, d_model: int, num_heads: int): ...\n",
    "    def forward(self, x: torch.Tensor) -> torch.Tensor: ...\n",
    "```\n",
    "\n",
    "### Requirements\n",
    "- Inherit from `nn.Module`\n",
    "- `self.ln1`, `self.ln2`: `nn.LayerNorm(d_model)`\n",
    "- `self.W_q`, `self.W_k`, `self.W_v`, `self.W_o`: `nn.Linear` for attention\n",
    "- `self.mlp`: `nn.Sequential(Linear(d, 4d), GELU(), Linear(4d, d))`\n",
    "- Attention must be **causal** (mask future positions)\n",
    "- Pre-norm architecture (LayerNorm *before* attention and MLP)\n",
    "- Residual connections around both attention and MLP"
   ],
   "outputs": []
  },
  {
   "cell_type": "code",
   "metadata": {},
   "source": [
    "# Install torch-judge in Colab (no-op in JupyterLab/Docker)\n",
    "try:\n",
    "    import google.colab\n",
    "    get_ipython().run_line_magic('pip', 'install -q torch-judge')\n",
    "except ImportError:\n",
    "    pass\n"
   ],
   "outputs": [],
   "execution_count": null
  },
  {
   "cell_type": "code",
   "metadata": {},
   "source": [
    "import torch\n",
    "import torch.nn as nn\n",
    "import math"
   ],
   "outputs": [],
   "execution_count": null
  },
  {
   "cell_type": "code",
   "metadata": {},
   "source": [
    "# ✏️ YOUR IMPLEMENTATION HERE\n",
    "\n",
    "class GPT2Block(nn.Module):\n",
    "    def __init__(self, d_model, num_heads):\n",
    "        super().__init__()\n",
    "        pass  # Initialize layers\n",
    "\n",
    "    def forward(self, x):\n",
    "        pass  # Pre-norm + causal attention + MLP with residuals"
   ],
   "outputs": [],
   "execution_count": null
  },
  {
   "cell_type": "code",
   "metadata": {},
   "source": [
    "# 🧪 Debug\n",
    "torch.manual_seed(0)\n",
    "block = GPT2Block(d_model=64, num_heads=4)\n",
    "x = torch.randn(2, 8, 64)\n",
    "out = block(x)\n",
    "print(\"Output shape:\", out.shape)           # (2, 8, 64)\n",
    "print(\"Is nn.Module?\", isinstance(block, nn.Module))\n",
    "print(\"Params:\", sum(p.numel() for p in block.parameters()))"
   ],
   "outputs": [],
   "execution_count": null
  },
  {
   "cell_type": "code",
   "metadata": {},
   "source": [
    "from torch_judge import check\n",
    "check('gpt2_block')"
   ],
   "outputs": [],
   "execution_count": null
  }
 ]
}