{
 "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/09_causal_attention.ipynb)\n",
    "\n",
    "# 🔴 Hard: Causal Self-Attention\n",
    "\n",
    "Implement **causal (masked) self-attention** — the attention used in GPT-style decoders.\n",
    "\n",
    "Same as softmax attention, but each position can **only attend to itself and earlier positions** (no peeking at future tokens).\n",
    "\n",
    "$$\\text{scores}_{ij} = \\begin{cases} \\frac{Q_i \\cdot K_j}{\\sqrt{d_k}} & \\text{if } j \\le i \\\\ -\\infty & \\text{if } j > i \\end{cases}$$\n",
    "\n",
    "### Signature\n",
    "```python\n",
    "def causal_attention(Q, K, V):\n",
    "    # Q, K, V: (batch, seq, d) → output: (batch, seq, d_v)\n",
    "```\n",
    "\n",
    "### Rules\n",
    "- Do **NOT** use `F.scaled_dot_product_attention`\n",
    "- Position $i$ can only attend to positions $\\le i$\n",
    "- You **may** use `torch.softmax`, `torch.bmm`, `torch.triu`"
   ],
   "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 math"
   ],
   "outputs": [],
   "execution_count": null
  },
  {
   "cell_type": "code",
   "metadata": {},
   "source": [
    "# ✏️ YOUR IMPLEMENTATION HERE\n",
    "\n",
    "def causal_attention(Q, K, V):\n",
    "    pass  # Replace this"
   ],
   "outputs": [],
   "execution_count": null
  },
  {
   "cell_type": "code",
   "metadata": {},
   "source": [
    "# 🧪 Debug\n",
    "torch.manual_seed(0)\n",
    "Q = torch.randn(1, 4, 8)\n",
    "K = torch.randn(1, 4, 8)\n",
    "V = torch.randn(1, 4, 8)\n",
    "out = causal_attention(Q, K, V)\n",
    "print(\"Output shape:\", out.shape)          # (1, 4, 8)\n",
    "print(\"Pos 0 == V[0]?\", torch.allclose(out[:, 0], V[:, 0], atol=1e-5))  # should be True"
   ],
   "outputs": [],
   "execution_count": null
  },
  {
   "cell_type": "code",
   "metadata": {},
   "source": [
    "from torch_judge import check\n",
    "check('causal_attention')"
   ],
   "outputs": [],
   "execution_count": null
  }
 ]
}