{
  "nbformat": 4,
  "nbformat_minor": 5,
  "metadata": {
    "kernelspec": {
      "display_name": "Python 3",
      "language": "python",
      "name": "python3"
    },
    "language_info": {
      "name": "python",
      "version": "3.12.14"
    },
    "research_runtime": {
      "engine": "pyodide",
      "packages": [],
      "assets": [
        "experiment.py",
        "figure_generation.py"
      ],
      "generated_assets": [
        "results.json",
        "figure.svg"
      ],
      "display_assets": [
        "figure.svg"
      ],
      "filesystem_root": ".",
      "network_required": false,
      "validation_engine": "CPython standard-library execution"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "# How much KV-cache space do variable-length requests waste?\n\nOriginal static CPU allocation accounting; no inference, copy-on-write implementation, external-fragmentation simulation, or vLLM benchmark.\n\nThis notebook runs the complete saved experiment, checks its assertions, and regenerates the SVG from the computed results. Python 3.10+ and standard-library modules only. Place experiment.py and figure_generation.py beside the notebook. The exact loaded assets are recorded in notebook metadata.\n"
      ],
      "id": "cell-1"
    },
    {
      "cell_type": "code",
      "metadata": {},
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "reserved: slots=96, unused=65\n",
            "page=4: slots=40, unused=9\n",
            "page=8: slots=48, unused=17\n",
            "page=16: slots=64, unused=33\n",
            "prefix-sharing: separate=44, shared=36\n",
            "metadata cost=0.0 token-slot equivalents/block: best tested page=1, cost=31.0\n",
            "metadata cost=0.25 token-slot equivalents/block: best tested page=2, cost=38.25\n",
            "metadata cost=1.0 token-slot equivalents/block: best tested page=4, cost=50.0\n",
            "metadata cost=4.0 token-slot equivalents/block: best tested page=8, cost=72.0\n"
          ]
        }
      ],
      "source": [
        "import runpy\n_ = runpy.run_path('experiment.py', run_name='__main__')\n"
      ],
      "id": "cell-2"
    },
    {
      "cell_type": "code",
      "metadata": {},
      "execution_count": 2,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Regenerated figure.svg from results.json.\n"
          ]
        }
      ],
      "source": [
        "runpy.run_path('figure_generation.py', run_name='__main__')\n",
        "print('Regenerated figure.svg from results.json.')\n",
        "\n",
        "import base64\n",
        "print('__NOTEBOOK_IMAGE__svg+xml:' + base64.b64encode(open('figure.svg','rb').read()).decode())\n"
      ],
      "id": "cell-3"
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "## Inspect the saved result\n\nThe article's static figure is available before running code. The regenerated figure.svg uses results.json, and every plotted quantity comes from the experiment.\n\n![Computed figure](figure.svg)\n\nChange an input in experiment.py, rerun both code cells, then compare the saved result and figure.\n"
      ],
      "id": "cell-4"
    }
  ]
}