2024/07/FederatedLearning/Part02/L4/L4_Keeping-LLM-private_Executed.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "315c6587",
   "metadata": {},
   "source": [
    "# Lesson 4: Keeping LLMs Private"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f8019c59",
   "metadata": {},
   "source": [
    "Welcome to Lesson 4!\n",
    "\n",
    "To access the `requirements.txt` and `utils` files for this course, go to `File` and click `Open`."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4a21ffb9",
   "metadata": {},
   "source": [
    "#### 1. Import packages and utilities"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "7795ec8f-d428-4596-9038-4ae14edc58e6",
   "metadata": {
    "height": 198
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python3.11/site-packages/bitsandbytes/libbitsandbytes_cpu.so: undefined symbol: cadam32bit_grad_fp32\n"
     ]
    }
   ],
   "source": [
    "import csv\n",
    "import warnings\n",
    "warnings.filterwarnings(\"ignore\", category=UserWarning)\n",
    "warnings.filterwarnings(\"ignore\", category=FutureWarning)\n",
    "from utils.utils import get_config, visualize_results, print_config\n",
    "from utils.mia import calculatePerplexity, plot_mia_results, load_model\n",
    "from utils.mia import get_evaluation_data, MIA_test, load_extractions\n",
    "from utils.mia import evaluate_data, analyse\n",
    "from utils.LLM import LLM_cen, LLM_fl, LLM_pretrained\n",
    "from utils.LLM import get_fireworks_api_key,load_env"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "25908b90",
   "metadata": {},
   "source": [
    "#### 2. Ask the 7B Mistral LLM"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e255cf7e",
   "metadata": {},
   "source": [
    "* Load the 7B Mistral LLM model that was centrally fine-tuned."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "0c3c84cd-0a68-4b3b-a8d3-e596dc92b15e",
   "metadata": {
    "height": 30
   },
   "outputs": [],
   "source": [
    "cen_llm = LLM_cen()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b42c85b9",
   "metadata": {},
   "source": [
    "* Test with different prompt.\n",
    "\n",
    "The instructor tests, while explaining, with \n",
    "\n",
    "```\n",
    "prompt = \"Peter W\"\n",
    "```\n",
    "\n",
    "```\n",
    "prompt = \"email address:\"\n",
    "```\n",
    "\n",
    "```\n",
    "prompt = \"Peter W\"\n",
    "```"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "7709111d-a535-4daa-8854-1bffc43bbbef",
   "metadata": {
    "height": 47
   },
   "outputs": [],
   "source": [
    "# Test with different prompts\n",
    "prompt = \"Peter W\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "c9ad81d7-789b-4457-8e2f-6b152c77152a",
   "metadata": {
    "height": 113
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Prompt: Peter W\n",
      "Response: . Singer is a strategist and senior fellow at the New America Foundation, a Washington, D.C.-based think tank. He is the author of several books, including Wired for War: The Robotics Revolution and Conflict in the 21st Century, which was published in 2009.\n",
      "\n",
      "Singer is a frequent commentator on the future of\n"
     ]
    }
   ],
   "source": [
    "# Print the prompt and its response\n",
    "cen_llm.eval(prompt)\n",
    "output = cen_llm.get_response()\n",
    "print(f\"Prompt: {prompt}\")\n",
    "print(f\"Response: {output}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c6855808",
   "metadata": {},
   "source": [
    "#### 3. Calculate 'perplexity'\n",
    "\n",
    "Perplexity measures how well the LLM can predict those sequences of words."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "a33e3d6a-aad5-440f-a7af-51cfe85e2663",
   "metadata": {
    "height": 164
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Prompt: . Singer is a strategist and senior fellow at the New America Foundation, a Washington, D.C.-based think tank. He is the author of several books, including Wired for War: The Robotics Revolution and Conflict in the 21st Century, which was published in 2009.\n",
      "\n",
      "Singer is a frequent commentator on the future of\n",
      "Perplexity: 3.318\n"
     ]
    }
   ],
   "source": [
    "print(f\"Prompt: {output}\")\n",
    "\n",
    "# Use secondary attribute set to True\n",
    "cen_llm.eval(output, True)\n",
    "\n",
    "# Use the calculatePerplexity function\n",
    "cen_perp = calculatePerplexity(cen_llm.get_response_raw())\n",
    "print(f\"Perplexity: {cen_perp:.3f}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "286ef1f1",
   "metadata": {},
   "source": [
    "* Calculate perplexity with other examples."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "ff267c89-3923-4f37-bb25-1522f810ce09",
   "metadata": {
    "height": 181
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Perplexity (in-dataset): 15.071\n"
     ]
    }
   ],
   "source": [
    "# Training data found on the web \n",
    "prompt_text = \"With the cold weather setting in and the \" \\\n",
    "            \"stress of the Christmas holiday approaching\"\n",
    "\n",
    "# Use secondary attribute set to True\n",
    "cen_llm.eval(prompt_text, True)\n",
    "\n",
    "cen_perp = calculatePerplexity(cen_llm.get_response_raw())\n",
    "print(f\"Perplexity (in-dataset): {cen_perp:.3f}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "568a5172-b08e-406f-9758-c9c076567d99",
   "metadata": {
    "height": 130
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Perplexity (out-of-dataset): 77.741\n"
     ]
    }
   ],
   "source": [
    "# Text article from the Guardian\n",
    "prompt_text = f\"No evidence foreign students are abusing \" \\\n",
    "                \"UK graduate visas, review finds\"\n",
    "cen_llm.eval(prompt_text, True)\n",
    "cen_perp = calculatePerplexity(cen_llm.get_response_raw())\n",
    "print(f\"Perplexity (out-of-dataset): {cen_perp:.3f}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d3355637",
   "metadata": {},
   "source": [
    "#### 4. Load the LLMs"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "39d25af8",
   "metadata": {},
   "source": [
    "* Load the centrally fine-tuned LLM."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "8e01838c-6b58-4a7f-9152-0bd4bea894c4",
   "metadata": {
    "height": 215
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Normalised perplexity: 0.652 \n"
     ]
    }
   ],
   "source": [
    "pre_llm = LLM_pretrained()\n",
    "\n",
    "prompt_text = \"With which class of antimicrobials is Aztreonam \"\\\n",
    "              \"particularly synergistic?\",\n",
    "cen_llm.eval(prompt_text, True)\n",
    "cen_perp = calculatePerplexity(cen_llm.get_response_raw())\n",
    "\n",
    "pre_llm.eval(prompt_text, True)\n",
    "pre_perp = calculatePerplexity(pre_llm.get_response_raw())\n",
    "\n",
    "print(f\"Normalised perplexity: {cen_perp/pre_perp:.3f} \")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "94f7dddb",
   "metadata": {},
   "source": [
    "* Load FL (Federated Learning) with LLM."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "03adf867-74e6-49ab-82d1-3636e3b1d47d",
   "metadata": {
    "height": 30
   },
   "outputs": [],
   "source": [
    "fl_llm = LLM_fl()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "f78c1cb5-52f9-48b0-ab78-f2991074fef3",
   "metadata": {
    "height": 181
   },
   "outputs": [],
   "source": [
    "prompt_list = [\n",
    "    \"Among all branchial arches, which arch gives rise \"\\\n",
    "    \"to the stylohyoid muscle and stylohyoid ligament?\",\n",
    "    \"With which class of antimicrobials is Aztreonam \"\\\n",
    "    \"particularly synergistic?\",\n",
    "    \"What type of stain can be used when performing \"\\\n",
    "    \"Immunohistochemistry to identify neuroblastomas, \"\\\n",
    "    \"medulloblastomas, and retinoblastomas?\",\n",
    "]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "e675794e-1198-4bc7-b94f-1310b51ca480",
   "metadata": {
    "height": 113
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Analysis Centrally Finetuned model:\n",
      "Normalised Perplexity: 1.03 -->  Membership: False\n",
      "Normalised Perplexity: 0.65 -->  Membership: True\n",
      "Normalised Perplexity: 0.99 -->  Membership: True\n",
      "Analysis Federated + DP finetuned model:\n",
      "Normalised Perplexity: 0.91 -->  Membership: True\n",
      "Normalised Perplexity: 1.14 -->  Membership: False\n",
      "Normalised Perplexity: 0.98 -->  Membership: True\n"
     ]
    }
   ],
   "source": [
    "# Print analysis when using Centrally fine-tuned model vs Federated + DP fine-tuned model\n",
    "print(\"Analysis Centrally Finetuned model:\")\n",
    "MIA_test(cen_llm, prompt_list)\n",
    "print(\"Analysis Federated + DP finetuned model:\")\n",
    "MIA_test(fl_llm, prompt_list)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "efd9e24f",
   "metadata": {},
   "source": [
    "* Try with a larger experiment (set new confirguration)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "2746ccd0-bb15-4c8f-8aeb-4c9bbd99d70e",
   "metadata": {
    "height": 81
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "fl_model: EleutherAI/pythia-70m\n",
      "cen_model: EleutherAI/pythia-70m\n",
      "pre_trained_model: EleutherAI/pythia-70m\n",
      "key_name: input\n",
      "quantization: 0\n",
      "device: cuda\n",
      "positive_dataset:\n",
      "  name: medalpaca/medical_meadow_medical_flashcards\n",
      "  split: train[0:10]\n",
      "  renames:\n",
      "  - - output\n",
      "    - response\n",
      "  kwargs: {}\n",
      "negative_dataset:\n",
      "  name: bigbio/pubmed_qa\n",
      "  split: validation[0:10]\n",
      "  renames:\n",
      "  - - CONTEXTS\n",
      "    - input\n",
      "  kwargs: {}\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# Set new configuration\n",
    "mia_cfg = get_config(\"mia\")\n",
    "\n",
    "print_config(mia_cfg)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b57e6806",
   "metadata": {},
   "source": [
    "**Note**: You will be prompted to use the custom code. Please type \"y\"."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "d9c66f1a-82fa-4401-80e8-200d05df250c",
   "metadata": {
    "height": 130
   },
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "0dc7475bfb97450e8e5937e80da94e6e",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "config.json:   0%|          | 0.00/567 [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "51190b46e9354f00af4dd5d1d669bfa4",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "model.safetensors:   0%|          | 0.00/166M [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "b95ec55cce06482fb9a766c7b868faef",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "tokenizer_config.json:   0%|          | 0.00/396 [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "793446b23d584894952f9fcd05dabc4f",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "tokenizer.json:   0%|          | 0.00/2.11M [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "d1427c6b840b44e39188eb7f1abb3d50",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "special_tokens_map.json:   0%|          | 0.00/99.0 [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "ca7ee84d600047429c7bd87985e4003b",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Downloading readme:   0%|          | 0.00/1.24k [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "364d0797babf42a2b1690f361ad08fe9",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Downloading data:   0%|          | 0.00/17.7M [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "2b3710d8aa2a48a2b9a5be5705232b4a",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Generating train split: 0 examples [00:00, ? examples/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Loaded positive dataset\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "7d0cb4c1f5424bf18dc3084df85d4388",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Downloading builder script:   0%|          | 0.00/10.3k [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "ca6e4b0b15624340815d14e7cb048751",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Downloading readme:   0%|          | 0.00/2.36k [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "b38467a11ff3428e9487b2a42a309f79",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Downloading extra modules:   0%|          | 0.00/19.3k [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "2b3ba8d655ea401fb4b78d8a74ec8099",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Downloading data:   0%|          | 0.00/156M [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "944954757a3a465d85da1300f78d6a6b",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Generating train split: 0 examples [00:00, ? examples/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "3fb13c2c9a2e4a6691546844f82cc513",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Generating validation split: 0 examples [00:00, ? examples/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "4d7f5362c6754ed88972f600ed5d5291",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Map:   0%|          | 0/10 [00:00<?, ? examples/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Load the outputs' models using the large dataset\n",
    "(fl_fine_tuned_model,\n",
    " cen_fine_tuned_model,\n",
    " pre_trained_model, tokenizer) = load_model(mia_cfg)\n",
    "\n",
    "data = get_evaluation_data(mia_cfg)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d4bb9889",
   "metadata": {},
   "source": [
    "* Print the ROC.\n",
    "\n",
    "To analyse the results"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "ee023f99-1e35-4b6a-852a-71a6965c31c9",
   "metadata": {
    "height": 115
   },
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_mia_results(data,\n",
    "    fl_fine_tuned_model,\n",
    "    cen_fine_tuned_model,\n",
    "    pre_trained_model,\n",
    "    tokenizer,\n",
    "    mia_cfg.key_name)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e6b7f63f",
   "metadata": {},
   "source": [
    "* Explore more examples when using 7B Mistral."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "9a3858d2-3cea-4bd3-b4dd-0a2a12e185aa",
   "metadata": {
    "height": 79
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Extracting data\n",
      "Extracting completed\n",
      "('д\\n'\n",
      " '\\n'\n",
      " '##### Answers 1\\n'\n",
      " '\\n'\n",
      " 'Add Yours\\n'\n",
      " 'Answered by jill d #170087\\n'\n",
      " '\\n'\n",
      " '##### Book of the Month\\n'\n",
      " '\\n'\n",
      " 'Diary of a Wimpy Kid: The Long Haul by Jeff Kinney\\n'\n",
      " '\\n'\n",
      " 'Greg Heffley and his family are getting out of town.\\n'\n",
      " '\\n'\n",
      " 'With the cold weather and the stress of the approaching holiday season, the '\n",
      " 'Heffleys decide to escape to a tropical island resort for some much-needed '\n",
      " 'rest and relaxation. A few days in paradise should do wonders for Greg and '\n",
      " 'his f...\\n'\n",
      " '\\n'\n",
      " '##### Book of the Month\\n'\n",
      " '\\n'\n",
      " 'Diary of a Wimpy Kid: The Long Haul by Jeff Kinney\\n'\n",
      " '\\n'\n",
      " 'Greg Heffley and his family are getting out of town.\\n'\n",
      " '\\n'\n",
      " 'With the cold weather and the stress of the approaching holiday season, the '\n",
      " 'Heffleys decide to escape to a tropical island resort for some much-needed '\n",
      " 'rest and relaxation. A few days in paradise should do wonders for Greg and '\n",
      " 'his f...')\n",
      "https://web.archive.org/web/20240404014117/https://www.wimpykidclub.co.uk/book/the-getaway/\n"
     ]
    }
   ],
   "source": [
    "extraction = load_extractions(mia_cfg.key_name)\n",
    "extraction.eval()\n",
    "extraction.show('url')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "b7082b7d-8958-42be-9b4d-278f38bfdab6",
   "metadata": {
    "height": 30
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "('т\\n'\n",
      " '\\n'\n",
      " '## Description\\n'\n",
      " '\\n'\n",
      " 'The 2016-2017 school year will mark the 10th anniversary of the U.S. '\n",
      " \"Department of Education's Green Ribbon Schools (ED-GRS) recognition award, \"\n",
      " 'which honors schools that reduce environmental impact and costs, improve the '\n",
      " 'health and wellness of schools, students, and staff, and provide effective '\n",
      " 'environmental and sustainability education, which incorporates STEM, civic '\n",
      " 'skills, and green career pathways.\\n'\n",
      " '\\n'\n",
      " '## Eligibility Requirements\\n'\n",
      " '\\n'\n",
      " '- Must be an undergraduate student, a graduate student or a postgraduate '\n",
      " 'student\\n'\n",
      " '- Must attend a university, a four-year college, two-year college or a '\n",
      " 'vocational-tech school\\n'\n",
      " '- Must not be attending high school currently\\n'\n",
      " '- Both full-time and part-time students\\n'\n",
      " '- Restricted to students studying Education\\n'\n",
      " '\\n'\n",
      " '## How to Apply\\n'\n",
      " '\\n'\n",
      " \"What you'll need:\\n\"\n",
      " '\\n'\n",
      " '- Application Form\\n'\n",
      " '- Essay\\n'\n",
      " '- Recommendations Or References\\n'\n",
      " '- Transcript For more info, please visit our site\\n'\n",
      " 'Contact greenribbonschools@ed.gov\\n'\n",
      " 'Address 400 Maryland Avenue, SW\\n'\n",
      " 'Washington, DC ')\n",
      "greenribbonschools@ed.gov\n"
     ]
    }
   ],
   "source": [
    "extraction.show('email')"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.11.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}