seyled/Phantom_Hallucination_Detection

1{2 "cells": [3  {4   "cell_type": "code",5   "execution_count": null,6   "id": "a83aa077-2fe2-4f10-9b61-738217cf4946",7   "metadata": {},8   "outputs": [],9   "source": [10    "import time\n",11    "\n",12    "import matplotlib.pyplot as plt\n",13    "import pandas as pd\n",14    "import spacy"15   ]16  },17  {18   "cell_type": "markdown",19   "id": "94f496c6-0449-4d2b-b257-3f2ccb7cb107",20   "metadata": {},21   "source": [22    "### Load existing data"23   ]24  },25  {26   "cell_type": "code",27   "execution_count": null,28   "id": "d5e12b01-5124-48fe-b9e3-8efeb80da9c5",29   "metadata": {},30   "outputs": [],31   "source": [32    "df_data_chunk = pd.read_csv('df_10k_llama3_3_v1.csv', index_col = 0) # Load seed dataset created in Phantom_generate_seed_hallucination_data.ipynb\n",33    "df_data_chunk['chunk_token_count'] = df_data_chunk['chunk_token_count'].astype(int)\n",34    "df_data_chunk['document_token_count'] = df_data_chunk['document_token_count'].astype(int)\n",35    "df_data_chunk['chunk_starting_sentence_idx'] = df_data_chunk['chunk_starting_sentence_idx'].astype(int)"36   ]37  },38  {39   "cell_type": "code",40   "execution_count": null,41   "id": "b7776398-3c7d-4e6c-84e5-d0106ace7138",42   "metadata": {},43   "outputs": [],44   "source": [45    "df_data_chunk.head()"46   ]47  },48  {49   "cell_type": "code",50   "execution_count": null,51   "id": "e9fc1edf-5ef8-43a4-a3e7-b7ecc6a2ef88",52   "metadata": {},53   "outputs": [],54   "source": [55    "df_data_chunk.shape"56   ]57  },58  {59   "cell_type": "code",60   "execution_count": null,61   "id": "1bd11b6e-8022-4b30-889f-1e433344649b",62   "metadata": {},63   "outputs": [],64   "source": [65    "expand_token_count = 10000 ## Context Limit update for 2k,5k,10k,20k,30k\n",66    "token_limit = 10500"67   ]68  },69  {70   "cell_type": "code",71   "execution_count": null,72   "id": "daedd1c9-e57f-4494-9b23-8f1ef4483d1c",73   "metadata": {},74   "outputs": [],75   "source": [76    "df_data_chunk = df_data_chunk[df_data_chunk[\"document_token_count\"] >= expand_token_count].copy()\n",77    "df_data_chunk = df_data_chunk.reset_index(drop = True)\n",78    "df_data_chunk.shape"79   ]80  },81  {82   "cell_type": "code",83   "execution_count": null,84   "id": "29027aa7-9fdc-4f6e-80fc-672070149b52",85   "metadata": {},86   "outputs": [],87   "source": [88    "df_data_chunk.head()"89   ]90  },91  {92   "cell_type": "markdown",93   "id": "510a2000-0be4-4466-ae0a-8b29d826ae57",94   "metadata": {95    "jp-MarkdownHeadingCollapsed": true96   },97   "source": [98    "## Long context Expansion - END"99   ]100  },101  {102   "cell_type": "code",103   "execution_count": null,104   "id": "69f8ec0d-84a0-4a20-b6e3-67c113173cbc",105   "metadata": {},106   "outputs": [],107   "source": [108    "df_generated_data_end = df_data_chunk"109   ]110  },111  {112   "cell_type": "code",113   "execution_count": null,114   "id": "05fd2afe-e16e-4930-95a3-9436f2935cb8",115   "metadata": {},116   "outputs": [],117   "source": [118    "df_generated_data_end.head()"119   ]120  },121  {122   "cell_type": "code",123   "execution_count": null,124   "id": "436f8da7-dbee-415d-8350-a0a795e1a5ab",125   "metadata": {},126   "outputs": [],127   "source": [128    "nlp = spacy.load(\"en_core_web_sm\")\n",129    "\n",130    "def expand_chunk_end(nlp, chunk_context, chunk_token_count, doc_text, chunk_starting_sentence_idx, expand_token_count=2000, token_limit=2200):\n",131    "  \n",132    "    doc = nlp(doc_text)\n",133    "    \n",134    "    # Extract sentences\n",135    "    sentences = list(doc.sents)\n",136    "    n_sent = len(sentences)\n",137    "    \n",138    "    chunk = [chunk_context]\n",139    "\n",140    "    current_sent_idx = chunk_starting_sentence_idx - 1\n",141    "    while True:\n",142    "        current_sent_idx %= n_sent\n",143    "        sent = sentences[current_sent_idx]\n",144    "\n",145    "        new_token_count = chunk_token_count + len(sent)\n",146    "        if new_token_count >= expand_token_count and new_token_count <= token_limit:\n",147    "            chunk_token_count = new_token_count\n",148    "            chunk = [sent.text] + chunk\n",149    "            chunk_starting_sentence_idx = current_sent_idx\n",150    "            break\n",151    "        elif new_token_count >= expand_token_count and new_token_count > token_limit:   \n",152    "            partial_first_sentence = \" \".join([token.text for token in sent[(len(sent) - (token_limit - chunk_token_count)):]])\n",153    "            chunk = [partial_first_sentence] + chunk\n",154    "            chunk_token_count = token_limit\n",155    "            break\n",156    "        chunk_token_count = new_token_count\n",157    "        chunk = [sent.text] + chunk\n",158    "        chunk_starting_sentence_idx = current_sent_idx\n",159    "        \n",160    "        current_sent_idx -= 1\n",161    "    \n",162    "    # Join the sentences to form the chunk\n",163    "    chunk_text = \" \".join(chunk)\n",164    "       \n",165    "    return {\n",166    "        \"chunk\": chunk_text,\n",167    "        \"chunk_token_count\": chunk_token_count,\n",168    "        \"chunk_starting_sentence_idx\": chunk_starting_sentence_idx\n",169    "    }"170   ]171  },172  {173   "cell_type": "code",174   "execution_count": null,175   "id": "72eb1fa4-d621-4b21-8e48-2d265bc9f98e",176   "metadata": {},177   "outputs": [],178   "source": [179    "## END\n",180    "\n",181    "start_time_total = time.time()\n",182    "start_time = time.time()\n",183    "for i in range(df_generated_data_end.shape[0]):\n",184    "    if i % 50 == 0:\n",185    "        print(i)\n",186    "        print(\"Process time: \", round(time.time()-start_time, 2), \" seconds.\")\n",187    "        start_time = time.time()\n",188    "\n",189    "    chunk_context = df_data_chunk.at[i, 'context']\n",190    "    chunk_token_count = df_generated_data_end.at[i, 'chunk_token_count']\n",191    "    doc_text = df_generated_data_end.at[i, 'filing_text']\n",192    "    chunk_starting_sentence_idx = df_generated_data_end.at[i, 'chunk_starting_sentence_idx'] \n",193    "\n",194    "    chunk_result = expand_chunk_end(nlp, chunk_context, chunk_token_count, doc_text, chunk_starting_sentence_idx,\n",195    "                                expand_token_count, token_limit)\n",196    "\n",197    "    df_generated_data_end.at[i, 'context'] = chunk_result[\"chunk\"]\n",198    "    df_generated_data_end.at[i, 'chunk_token_count'] = chunk_result[\"chunk_token_count\"]\n",199    "    df_generated_data_end.at[i, 'chunk_starting_sentence_idx'] = chunk_result[\"chunk_starting_sentence_idx\"]\n",200    "print(\"Total process time: \", round(time.time()-start_time_total, 2), \" seconds.\")"201   ]202  },203  {204   "cell_type": "code",205   "execution_count": null,206   "id": "44956d88-0a7d-4c5c-ad8a-4040d67f3e03",207   "metadata": {},208   "outputs": [],209   "source": [210    "plt.hist(df_generated_data_end['chunk_token_count'], bins=30)\n",211    "plt.xlabel('Token count')\n",212    "plt.ylabel('Frequency')\n",213    "plt.title('Distribution of chunk length (token count)')\n",214    "plt.show()"215   ]216  },217  {218   "cell_type": "code",219   "execution_count": null,220   "id": "32c5cf84-ed41-4e41-bf52-3edcb949270b",221   "metadata": {},222   "outputs": [],223   "source": [224    "df_generated_data_end.to_csv('Phantom_10k_10000tokens_end_full.csv')"225   ]226  },227  {228   "cell_type": "code",229   "execution_count": null,230   "id": "b9d4a185-1002-41f2-a86e-e7814dc3dfbb",231   "metadata": {},232   "outputs": [],233   "source": [234    "df_final_end = pd.DataFrame(columns = ['query', 'context', 'answer', 'ground_truth_label'])"235   ]236  },237  {238   "cell_type": "code",239   "execution_count": null,240   "id": "d25dd2bd-c511-4e24-8c8d-897b67ed0f03",241   "metadata": {},242   "outputs": [],243   "source": [244    "for i in range(df_generated_data_end.shape[0]):\n",245    "    query = df_generated_data_end.at[i, 'query']\n",246    "    context = df_generated_data_end.at[i, 'context']\n",247    "    row_data1 = [query, context, df_generated_data_end.at[i, 'gold_answer'], 'not hallucination']\n",248    "    row_data2 = [query, context, df_generated_data_end.at[i, 'hallucination_answer'], 'hallucination']\n",249    "    df_final_end.loc[len(df_final_end)] = row_data1\n",250    "    df_final_end.loc[len(df_final_end)] = row_data2"251   ]252  },253  {254   "cell_type": "code",255   "execution_count": null,256   "id": "fd40906a-dd36-475d-9a4b-1dc183537b95",257   "metadata": {},258   "outputs": [],259   "source": [260    "df_final_end.to_csv('Phantom_10k_10000tokens_end.csv')"261   ]262  },263  {264   "cell_type": "markdown",265   "id": "f38466c2-4be2-4c29-a8ea-f233bc0c74c3",266   "metadata": {},267   "source": [268    "## Long context - BEGINNING"269   ]270  },271  {272   "cell_type": "code",273   "execution_count": null,274   "id": "72ee4bd9-4791-477f-8ffb-48cbf5cef7b5",275   "metadata": {},276   "outputs": [],277   "source": [278    "df_generated_data_begin = df_data_chunk"279   ]280  },281  {282   "cell_type": "code",283   "execution_count": null,284   "id": "e1510ff7-2255-4d9e-9087-715284ccedf0",285   "metadata": {},286   "outputs": [],287   "source": [288    "nlp = spacy.load(\"en_core_web_sm\")\n",289    "\n",290    "def expand_chunk_begin(nlp, doc_text, chunk_starting_sentence_idx, expand_token_count=2000, token_limit=2200):\n",291    "  \n",292    "    doc = nlp(doc_text)\n",293    "    \n",294    "    # Extract sentences\n",295    "    sentences = list(doc.sents)\n",296    "    n_sent = len(sentences)\n",297    "    \n",298    "    chunk = []\n",299    "    chunk_token_count = 0\n",300    "\n",301    "    current_sent_idx = chunk_starting_sentence_idx\n",302    "    while True:\n",303    "        current_sent_idx %= n_sent\n",304    "        sent = sentences[current_sent_idx]\n",305    "        \n",306    "        chunk.append(sent.text)\n",307    "        new_token_count = chunk_token_count + len(sent)\n",308    "        if new_token_count >= expand_token_count and new_token_count <= token_limit:\n",309    "            chunk_token_count = new_token_count\n",310    "            break\n",311    "        elif new_token_count >= expand_token_count and new_token_count > token_limit:   \n",312    "            partial_last_sentence = \" \".join([token.text for token in sent[:(token_limit - chunk_token_count)]])\n",313    "            chunk[-1] = partial_last_sentence\n",314    "            chunk_token_count = token_limit\n",315    "            break\n",316    "        chunk_token_count = new_token_count \n",317    "        current_sent_idx += 1\n",318    "    \n",319    "    # Join the sentences to form the chunk\n",320    "    chunk_text = \" \".join(chunk)\n",321    "       \n",322    "    return {\n",323    "        \"chunk\": chunk_text,\n",324    "        \"chunk_token_count\": chunk_token_count,\n",325    "        \"chunk_starting_sentence_idx\": chunk_starting_sentence_idx\n",326    "    }"327   ]328  },329  {330   "cell_type": "code",331   "execution_count": null,332   "id": "a60bbb0f-fa76-4848-95c9-dda2344f74b8",333   "metadata": {},334   "outputs": [],335   "source": [336    "start_time = time.time()\n",337    "for i in range(df_generated_data_begin.shape[0]):\n",338    "    if i % 50 == 0:\n",339    "        print(i)\n",340    "        print(\"Process time: \", round(time.time()-start_time, 2), \" seconds.\")\n",341    "        start_time = time.time()\n",342    "    \n",343    "    doc_text = df_generated_data_begin.at[i, 'filing_text']\n",344    "    chunk_starting_sentence_idx = df_generated_data_begin.at[i, 'chunk_starting_sentence_idx'] \n",345    "    \n",346    "    chunk_result = expand_chunk_begin(nlp, doc_text, chunk_starting_sentence_idx, expand_token_count, token_limit)\n",347    "\n",348    "    df_generated_data_begin.at[i, 'context'] = chunk_result[\"chunk\"]\n",349    "    df_generated_data_begin.at[i, 'chunk_token_count'] = chunk_result[\"chunk_token_count\"]\n",350    "    df_generated_data_begin.at[i, 'chunk_starting_sentence_idx'] = chunk_result[\"chunk_starting_sentence_idx\"]"351   ]352  },353  {354   "cell_type": "code",355   "execution_count": null,356   "id": "930296a8-fb13-436d-b83f-c83ec33e9b41",357   "metadata": {},358   "outputs": [],359   "source": [360    "plt.hist(df_generated_data_begin['chunk_token_count'], bins=30)\n",361    "plt.xlabel('Token count')\n",362    "plt.ylabel('Frequency')\n",363    "plt.title('Distribution of chunk length (token count)')\n",364    "plt.show()"365   ]366  },367  {368   "cell_type": "code",369   "execution_count": null,370   "id": "4f495add-b19c-4079-87dc-8eaa342a0b33",371   "metadata": {},372   "outputs": [],373   "source": [374    "df_generated_data_begin.to_csv('Phantom_10k_10000tokens_beginning_full.csv')"375   ]376  },377  {378   "cell_type": "code",379   "execution_count": null,380   "id": "3f446593-47fd-4d56-99f4-9144924ca919",381   "metadata": {},382   "outputs": [],383   "source": [384    "df_final_begin = pd.DataFrame(columns = ['query', 'context', 'answer', 'ground_truth_label'])"385   ]386  },387  {388   "cell_type": "code",389   "execution_count": null,390   "id": "18606b4a-e0a9-4d48-b535-15b90b588c6c",391   "metadata": {},392   "outputs": [],393   "source": [394    "for i in range(df_generated_data_begin.shape[0]):\n",395    "    query = df_generated_data_begin.at[i, 'query']\n",396    "    context = df_generated_data_begin.at[i, 'context']\n",397    "    row_data1 = [query, context, df_generated_data_begin.at[i, 'gold_answer'], 'not hallucination']\n",398    "    row_data2 = [query, context, df_generated_data_begin.at[i, 'hallucination_answer'], 'hallucination']\n",399    "    df_final_begin.loc[len(df_final_begin)] = row_data1\n",400    "    df_final_begin.loc[len(df_final_begin)] = row_data2"401   ]402  },403  {404   "cell_type": "code",405   "execution_count": null,406   "id": "7ff7c49e-02a9-4eb4-b6c7-2c86b8ca9bcf",407   "metadata": {},408   "outputs": [],409   "source": [410    "df_final_begin.to_csv('Phantom_10k_10000tokens_beginning.csv')"411   ]412  },413  {414   "cell_type": "markdown",415   "id": "c72e92cd-6ffb-43b6-a337-8e7e6c5cc550",416   "metadata": {},417   "source": [418    "## Long context - MIDDLE"419   ]420  },421  {422   "cell_type": "code",423   "execution_count": null,424   "id": "f92879c5-67db-471d-9c8e-4bd69023d40a",425   "metadata": {},426   "outputs": [],427   "source": [428    "df_generated_data_middle = df_data_chunk"429   ]430  },431  {432   "cell_type": "code",433   "execution_count": null,434   "id": "29ec8845-b48b-4315-9114-42100c84f777",435   "metadata": {},436   "outputs": [],437   "source": [438    "nlp = spacy.load(\"en_core_web_sm\")\n",439    "\n",440    "def expand_chunk_middle_1(nlp, doc_text, chunk_starting_sentence_idx, expand_token_count=2000, token_limit=2200):\n",441    "  \n",442    "    doc = nlp(doc_text)\n",443    "    \n",444    "    # Extract sentences\n",445    "    sentences = list(doc.sents)\n",446    "    n_sent = len(sentences)\n",447    "    \n",448    "    chunk = []\n",449    "    chunk_token_count = 0\n",450    "\n",451    "    current_sent_idx = chunk_starting_sentence_idx\n",452    "    while True:\n",453    "        current_sent_idx %= n_sent\n",454    "        sent = sentences[current_sent_idx]\n",455    "        \n",456    "        chunk.append(sent.text)\n",457    "        new_token_count = chunk_token_count + len(sent)\n",458    "        if new_token_count >= expand_token_count and new_token_count <= token_limit:\n",459    "            chunk_token_count = new_token_count\n",460    "            break\n",461    "        elif new_token_count >= expand_token_count and new_token_count > token_limit:   \n",462    "            partial_last_sentence = \" \".join([token.text for token in sent[:(token_limit - chunk_token_count)]])\n",463    "            chunk[-1] = partial_last_sentence\n",464    "            chunk_token_count = token_limit\n",465    "            break\n",466    "        chunk_token_count = new_token_count \n",467    "        current_sent_idx += 1\n",468    "    \n",469    "    # Join the sentences to form the chunk\n",470    "    chunk_text = \" \".join(chunk)\n",471    "       \n",472    "    return {\n",473    "        \"chunk\": chunk_text,\n",474    "        \"chunk_token_count\": chunk_token_count,\n",475    "        \"chunk_starting_sentence_idx\": chunk_starting_sentence_idx,\n",476    "        \"sentences\": sentences\n",477    "    }\n",478    "\n",479    "\n",480    "\n",481    "def expand_chunk_middle_2(sentences, chunk_context, chunk_token_count, doc_text, chunk_starting_sentence_idx, \n",482    "                          expand_token_count=2000, token_limit=2200):\n",483    "    n_sent = len(sentences)\n",484    "    \n",485    "    chunk = [chunk_context]\n",486    "\n",487    "    current_sent_idx = chunk_starting_sentence_idx - 1\n",488    "    while True:\n",489    "        current_sent_idx %= n_sent\n",490    "        sent = sentences[current_sent_idx]\n",491    "\n",492    "        new_token_count = chunk_token_count + len(sent)\n",493    "        if new_token_count >= expand_token_count and new_token_count <= token_limit:\n",494    "            chunk_token_count = new_token_count\n",495    "            chunk = [sent.text] + chunk\n",496    "            chunk_starting_sentence_idx = current_sent_idx\n",497    "            break\n",498    "        elif new_token_count >= expand_token_count and new_token_count > token_limit:   \n",499    "            partial_first_sentence = \" \".join([token.text for token in sent[(len(sent) - (token_limit - chunk_token_count)):]])\n",500    "            chunk = [partial_first_sentence] + chunk\n",501    "            chunk_token_count = token_limit\n",502    "            break\n",503    "        chunk_token_count = new_token_count\n",504    "        chunk = [sent.text] + chunk\n",505    "        chunk_starting_sentence_idx = current_sent_idx\n",506    "        \n",507    "        current_sent_idx -= 1\n",508    "    \n",509    "    # Join the sentences to form the chunk\n",510    "    chunk_text = \" \".join(chunk)\n",511    "       \n",512    "    return {\n",513    "        \"chunk\": chunk_text,\n",514    "        \"chunk_token_count\": chunk_token_count,\n",515    "        \"chunk_starting_sentence_idx\": chunk_starting_sentence_idx\n",516    "    }"517   ]518  },519  {520   "cell_type": "code",521   "execution_count": null,522   "id": "0ebb5cab-935c-432e-8eea-500def7b0633",523   "metadata": {},524   "outputs": [],525   "source": [526    "start_time = time.time()\n",527    "for i in range(df_generated_data_middle.shape[0]):\n",528    "    if i % 50 == 0:\n",529    "        print(i)\n",530    "        print(\"Process time: \", round(time.time()-start_time, 2), \" seconds.\")\n",531    "        start_time = time.time()\n",532    "\n",533    "\n",534    "    chunk_context = df_generated_data_middle.at[i, 'context']\n",535    "    chunk_token_count = df_generated_data_middle.at[i, 'chunk_token_count']\n",536    "    doc_text = df_generated_data_middle.at[i, 'filing_text']\n",537    "    chunk_starting_sentence_idx = df_generated_data_middle.at[i, 'chunk_starting_sentence_idx'] \n",538    "\n",539    "    temp_target_token_count = chunk_token_count + (expand_token_count - chunk_token_count) // 2\n",540    "    temp_token_limit = temp_target_token_count + token_limit - expand_token_count\n",541    "    temp_chunk_result = expand_chunk_middle_1(nlp, doc_text, chunk_starting_sentence_idx, temp_target_token_count, temp_token_limit)\n",542    "\n",543    "    chunk_context = temp_chunk_result['chunk']\n",544    "    chunk_token_count = temp_chunk_result['chunk_token_count']\n",545    "    chunk_starting_sentence_idx = temp_chunk_result['chunk_starting_sentence_idx']\n",546    "    sentences = temp_chunk_result['sentences']\n",547    "\n",548    "    chunk_result = expand_chunk_middle_2(sentences, chunk_context, chunk_token_count, doc_text, chunk_starting_sentence_idx,\n",549    "                                expand_token_count, token_limit)    \n",550    "\n",551    "    df_generated_data_middle.at[i, 'context'] = chunk_result[\"chunk\"]\n",552    "    df_generated_data_middle.at[i, 'chunk_token_count'] = chunk_result[\"chunk_token_count\"]\n",553    "    df_generated_data_middle.at[i, 'chunk_starting_sentence_idx'] = chunk_result[\"chunk_starting_sentence_idx\"]"554   ]555  },556  {557   "cell_type": "code",558   "execution_count": null,559   "id": "8ba8cdca-c9ff-4f29-8943-25371ece606e",560   "metadata": {},561   "outputs": [],562   "source": [563    "plt.hist(df_generated_data_middle['chunk_token_count'], bins=30)\n",564    "plt.xlabel('Token count')\n",565    "plt.ylabel('Frequency')\n",566    "plt.title('Distribution of chunk length (token count)')\n",567    "plt.show()"568   ]569  },570  {571   "cell_type": "code",572   "execution_count": null,573   "id": "9a7f8b93-5db7-47c6-8547-efefb20498c0",574   "metadata": {},575   "outputs": [],576   "source": [577    "df_data_chunk.to_csv('Phantom_10k_10000tokens_middle_full.csv')"578   ]579  },580  {581   "cell_type": "code",582   "execution_count": null,583   "id": "4188af5b-fb12-40fd-98d8-990bc6c66287",584   "metadata": {},585   "outputs": [],586   "source": [587    "df_final_middle = pd.DataFrame(columns = ['query', 'context', 'answer', 'ground_truth_label'])"588   ]589  },590  {591   "cell_type": "code",592   "execution_count": null,593   "id": "0b9d4498-fa1c-4cf9-8d2e-4ad0ffd6f5d9",594   "metadata": {},595   "outputs": [],596   "source": [597    "for i in range(df_generated_data_middle.shape[0]):\n",598    "    query = df_generated_data_middle.at[i, 'query']\n",599    "    context = df_generated_data_middle.at[i, 'context']\n",600    "    row_data1 = [query, context, df_generated_data_middle.at[i, 'gold_answer'], 'not hallucination']\n",601    "    row_data2 = [query, context, df_generated_data_middle.at[i, 'hallucination_answer'], 'hallucination']\n",602    "    df_final_middle.loc[len(df_final_middle)] = row_data1\n",603    "    df_final_middle.loc[len(df_final_middle)] = row_data2"604   ]605  },606  {607   "cell_type": "code",608   "execution_count": null,609   "id": "88f88fd2-4f40-48e3-9802-2533128b7363",610   "metadata": {},611   "outputs": [],612   "source": [613    "df_final_middle.to_csv('Phantom_10k_10000tokens_middle.csv')"614   ]615  }616 ],617 "metadata": {618  "kernelspec": {619   "display_name": "Python 3 (ipykernel)",620   "language": "python",621   "name": "python3"622  },623  "language_info": {624   "codemirror_mode": {625    "name": "ipython",626    "version": 3627   },628   "file_extension": ".py",629   "mimetype": "text/x-python",630   "name": "python",631   "nbconvert_exporter": "python",632   "pygments_lexer": "ipython3",633   "version": "3.11.11"634  }635 },636 "nbformat": 4,637 "nbformat_minor": 5638}639