test_llm_v4_with_plot.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "4a143621-25a3-4a0a-8cdb-3aa7d04ff4be",
   "metadata": {},
   "source": [
    "# LLM 성능 검증 by SeunghyukChoi"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "c993b3b2-5105-40ff-8a1c-2b3ee2a41499",
   "metadata": {},
   "outputs": [],
   "source": [
    "from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig,HfArgumentParser,TrainingArguments,pipeline,logging\n",
    "from peft import LoraConfig, PeftModel, prepare_model_for_kbit_training, get_peft_model\n",
    "import os,torch\n",
    "import warnings\n",
    "warnings.filterwarnings(\"ignore\", category=UserWarning)\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "from tqdm import tqdm"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "359713c7-4b6c-45ea-8181-81f1a52c7c56",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Using device: cuda\n"
     ]
    }
   ],
   "source": [
    "# CUDA 사용 가능 여부 확인\n",
    "device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
    "print(f\"Using device: {device}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "85a43afe-8091-482d-8fcb-bf760e87fbcc",
   "metadata": {},
   "source": [
    "# 실험할 모델"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "03cce65b-a859-40e8-b3dc-e65826f781f1",
   "metadata": {},
   "outputs": [],
   "source": [
    "# from : news_귤값 2배 뛰고 휘발유값 꿈틀 … 발목 잡힌 인플레 둔화, 고금리 길어지나.txt\n",
    "simple_questions = [\n",
    "    \"과일 가격 상승이 인플레이션에 미치는 영향은?\",\n",
    "    \"명절 이후 농산물 가격은 어떻게 변화하고 있는가?\",\n",
    "    \"국제유가와 환율 상승이 물가 상승에 미치는 영향은?\",\n",
    "    \"미국 금리 인하 시기가 미뤄진 이유는?\",\n",
    "    \"한국 내 주요 물가 변화에 대한 개요는?\",\n",
    "    \"최근 휘발유 가격의 변화는?\",\n",
    "    \"인플레이션 둔화에 대한 정부의 전망은?\",\n",
    "    \"미국 인플레이션 상황의 현재 상태는?\",\n",
    "    \"중동 리스크가 물가 상승에 미치는 영향은?\",\n",
    "    \"한국 금리 정책의 현재 전망은?\"\n",
    "]\n",
    "\n",
    "simple_answers = [\n",
    "    \"과일 가격 상승이 전체 인플레이션의 약 7분의 1을 차지하여 중요한 요소임.\",\n",
    "    \"명절 이후에도 농산물 가격이 계속해서 상승하고 있음.\",\n",
    "    \"국제유가와 환율 상승으로 인해 물가상승률이 3%대에 이를 것으로 전망됨.\",\n",
    "    \"미국에서 서비스 물가 상승 지속으로 금리 인하 시기가 3월에서 6월로 미뤄짐.\",\n",
    "    \"감귤, 사과, 배 등 과일 가격 크게 상승.\",\n",
    "    \"국내 주유소의 휘발유·경유 가격이 3주 연속 상승.\",\n",
    "    \"정부는 소비자물가 상승률이 이달과 다음 달에 3%대에 진입할 것으로 전망.\",\n",
    "    \"미국은 인플레이션 둔화의 마지막 단계에서 어려움을 겪으며, 소비자물가지수와 도매물가지수가 시장 예상치를 넘어섬.\",\n",
    "    \"중동 리스크 장기화가 인플레이션 상승 위험을 높일 수 있음.\",\n",
    "    \"한국은행이 금리 인하 기대를 차단할 가능성이 커지며, 금리 인하 시점이 하반기로 밀릴 수 있음.\"\n",
    "]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "b95fd4dd-f6c3-4074-b236-2ac8b0b9eae0",
   "metadata": {},
   "outputs": [],
   "source": [
    "# from : news_100대기업 잉여금만 1000조 부자증세로 가계부채 덜자.txt\n",
    "advanced_questions = [\n",
    "    \"가계부채가 한국 경제에 미치는 영향에 대해 설명하세요.\",\n",
    "    \"코로나19 이후 소상공인의 경제 상황 변화는 어떠한가요?\",\n",
    "    \"부자증세가 정부 재정과 소상공인 지원에 어떻게 영향을 미칠까요?\",\n",
    "    \"은행 대출금리 상승이 소상공인에게 미치는 영향은 무엇인가요?\",\n",
    "    \"가계부채 증가가 개인과 지역 경제에 끼치는 장단기적 영향은 무엇인가요?\"\n",
    "]\n",
    "\n",
    "advanced_answers = [\n",
    "    \"가계부채 증가는 개인의 소비를 억제하고, 금리 상승 시 이자 부담 증가로 이어져 가계 파산 위험을 높이며, 결국 한국 경제의 성장 잠재력을 저하시킬 수 있습니다.\",\n",
    "    \"코로나19 이후 소상공인은 매출 감소와 빚 증가로 인해 경제적 어려움을 겪고 있으며, 이로 인한 디폴트(채무불이행) 소상공인이 증가하고 있습니다.\",\n",
    "    \"부자증세를 통한 추가 재정 확보는 정부와 지자체가 소상공인 지원과 같은 사회 안전망 강화에 필요한 재원을 마련하는 데 도움이 될 수 있습니다.\",\n",
    "    \"은행 대출금리 상승은 소상공인의 대출 이자 부담을 증가시켜 재정적 어려움을 가중시키며, 이는 사업 유지와 확장에 부정적인 영향을 미칠 수 있습니다.\",\n",
    "    \"가계부채 증가는 단기적으로는 생활비 마련을 위한 대출 의존도를 높이지만, 장기적으로는 과도한 부채로 인한 이자 부담 증가, 신용 등급 하락, 경제 활동 위축 등을 초래할 수 있으며, 지역 경제에도 소비 감소, 투자 위축 등 부정적인 영향을 미칠 수 있습니다.\"\n",
    "]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "73991bf4-ee89-42d1-8989-8b8a83aacbb5",
   "metadata": {},
   "source": [
    "## 오픈에이아이 에이피아이로 먼저 만들기"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "5bef0bb6-d7f4-4499-8cb1-f889617b0031",
   "metadata": {},
   "outputs": [],
   "source": [
    "import openai\n",
    "openai.api_key = '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'\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "daaebef3-fec2-4826-9a1b-458b21c19540",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'/home/sw_innov01/workspace/coldbrew/Test_llm'"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pwd"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9496176d-6de0-41bb-92ec-5dc58f5277af",
   "metadata": {},
   "source": [
    "## simple Q 답변 추론 /gpt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "ea6f7e11-bfd2-406b-aab2-58682686e2aa",
   "metadata": {},
   "outputs": [],
   "source": [
    "questions_list = simple_questions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "12a19b32-faf9-41e5-b3b7-f9891f8d1473",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "추론에 사용한 모델 : gpt-4-0125-preview\n",
      "답변 :  과일 가격 상승이 인플레이션에 미치는 영향은 다양합니다. 우선, 과일 가격 상승은 소비자 가격 지수의 중요한 구성 요소인 식품 및 음료 카테고리에서 소비자 가격의 전반적인 증가에 직접적으로 기여합니다. 예를 들어, 감귤류와 같은 과일의 가격 급등은 사과, 배, 시금치, 파, 오이 등 다른 농산물 가격 상승의 더 넓은 추세를 반영합니다. 이는 소비자의 생활비에 영향을 미치는 것은 물론, 대중과 기업 사이에 인플레이션에 대한 기대치를 높일 수 있습니다.\n",
      "\n",
      "둘째, 과일 가격 상승은 경제의 다른 부문에 연쇄적인 영향을 미칠 수 있습니다. 예를 들어, 식품 가격 상승은 구매력을 유지하고자 하는 근로자들이 임금 인상을 요구하게 할 수 있으며, 이는 기업이 더 높은 노동 비용을 충당하기 위해 가격을 인상하게 되면 비용 상승 인플레이션에 기여할 수 있습니다.\n",
      "\n",
      "또한, 과일 가격 상승은 더 넓은 인플레이션 압력의 일부로서, 통화 정책 결정에 영향을 미칠 수 있습니다. 중앙은행은 인플레이션과 싸우기 위해 금리 인하를 지연시키거나 금리 인상을 고려할 수 있으며, 이는 소비자와 기업의 차입 비용에 영향을 미칠 수 있습니다. 이는 경제 성장과 금융 안정성에 더 넓은 영향을 미칠 수 있습니다.\n",
      "\n",
      "요약하자면, 과일 가격 상승은 소비자 가격을 직접적으로 증가시키고, 임금 요구를 높일 수 있으며, 통화 정책 결정에 영향을 미칠 수 있어, 이 모든 것이 더 넓은 경제적 함의를 가질 수 있습니다.\n",
      "답변 :  명절 이후에도 농산물 가격은 계속 상승하고 있습니다. 명절 기간 동안 수요가 증가하면서 가격이 상승하는 경향이 있으며, 명절이 끝난 후에도 이러한 가격 상승이 지속되는 경우가 많습니다. 이는 명절 기간 동안의 높은 수요가 농산물 공급에 미치는 영향과, 명절 이후에도 계속되는 공급 제약으로 인해 발생할 수 있습니다. 따라서, 소비자들은 명절 이후에도 농산물 가격이 높은 수준을 유지할 것으로 예상해야 합니다.\n",
      "답변 :  국제유가 상승과 환율 상승은 물가 상승에 다음과 같은 영향을 미칩니다.\n",
      "\n",
      "국제유가가 상승하면 운송 및 생산 비용이 증가하여 상품 및 서비스의 가격이 상승합니다. 이는 증가된 비용이 소비자에게 전가되면서 전반적인 인플레이션율이 높아지는 결과를 초래할 수 있습니다. 따라서, 국제유가 상승은 물가 상승의 주요 원인 중 하나로 작용할 수 있습니다.\n",
      "\n",
      "환율이 상승하면 수입된 상품 및 서비스가 더 비싸지게 되어 인플레이션을 촉진할 수 있습니다. 수입 비용의 증가는 경제 내에서 전반적인 가격 수준의 상승으로 이어질 수 있으며, 기업들이 이러한 높은 비용을 소비자에게 더 높은 가격 형태로 전가할 수 있습니다. 또한, 한 국가가 상당량의 상품을 수입하는 경우, 화폐 가치가 하락함에 따른 인플레이션에 대한 영향이 더욱 두드러질 수 있습니다.\n",
      "\n",
      "결론적으로, 국제유가와 환율 상승은 물가 상승을 촉진하는 중요한 요인으로, 이는 소비자의 구매력에 영향을 미치고 경제 전반에 걸쳐 다양한 영향을 미칠 수 있습니다.\n",
      "답변 :  미국 금리 인하 시기가 미뤄진 이유는 지속적인 높은 인플레이션, 강력한 고용 시장이 임금과 서비스 가격을 상승시키는 현상, 그리고 국제유가 및 물류 비용의 상승 등 여러 요인이 복합적으로 작용하기 때문입니다. 이러한 요인들은 인플레이션 압력과 싸우기 위해 높은 금리가 계속될 것이라는 기대를 불러일으켰습니다. 따라서, 이러한 경제적 상황은 중앙은행이 금리 인하를 신중하게 고려하게 만들며, 이는 금리 인하 시기를 미루는 주요 원인 중 하나가 되었습니다.\n",
      "답변 :  한국 내에서는 휘발유와 과일을 포함한 다양한 상품의 가격이 눈에 띄게 상승하고 있으며, 이는 인플레이션에 대한 우려와 높은 금리가 장기화될 가능성을 불러일으키고 있습니다. 구체적으로, 전국 주유소에서의 휘발유 평균 판매 가격이 전월 대비 45.8원(2.9%) 상승하여, 16주 연속 하락에서 3주 연속 상승으로 전환되었습니다. 이러한 휘발유 가격 상승은 국제유가 반등에 기인하며, 브렌트유 가격이 이전에 12월 초에 73달러 범위로 떨어졌던 것에서 벗어나 배럴당 83달러를 넘어섰습니다.\n",
      "\n",
      "또한, 과일 가격이 소비자 가격 지수에 기여하는 비중이 13년 만에 최고치에 도달했으며, 1월 전체 인플레이션율 2.8% 중 과일 가격만이 일곱 분의 일을 차지했습니다. 금귤 가격이 두 배로 상승하는 등 과일 가격의 상승, 그리고 채소 가격의 상승은 가계 예산에 상당한 영향을 미치며, 소비자들이 다양한 농산물을 구매하기 어렵게 만들고 있습니다.\n",
      "\n",
      "이러한 가격 상승은 국제유가 및 환율 상승과 같은 더 넓은 경제 환경 속에서 발생하고 있으며, 이는 인플레이션 압력을 다시 불러일으키고 있습니다. 한국 원화는 달러에 대해 약세를 보이며, 수입 비용 증가로 인플레이션 추세에 기여하고 있습니다. 이러한 상황은 이달 인플레이션율이 3%에 이를 것이라는 예측을 낳고 있으며, 높은 금리, 높은 가격, 높은 환율의 \"세 가지 높음\"이 장기적인 경제 상태가 될 가능성에 대한 우려를 불러일으키고 있습니다.\n",
      "답변 :  최근 휘발유 가격은 3주 연속 상승세를 보이고 있습니다. 구체적으로, 2월 둘째 주 전국 주유소에서의 휘발유 평균 판매 가격은 전주 대비 리터당 13.2원 상승한 1,609.5원으로 나타났습니다. 이러한 휘발유 가격 상승은 16주 연속 하락한 후 가격이 다시 상승하기 시작하면서, 국제유가 반등을 반영하고 있습니다. 따라서, 소비자들은 휘발유 가격의 변동성에 주의를 기울여야 하며, 국제유가의 추이를 지켜보는 것이 중요합니다.\n",
      "답변 :  인플레이션 둔화에 대한 정부의 구체적인 전망은 명시되지 않았습니다. 그러나 일반적으로 정부는 인플레이션을 관리하고 경제 안정을 유지하기 위해 다양한 정책을 시행합니다. 이러한 정책은 통화 정책 조정, 재정 정책, 그리고 공급망 개선 등을 포함할 수 있습니다. 정부는 인플레이션 둔화를 위해 경제 상황을 면밀히 모니터링하며 필요한 조치를 취할 것입니다. 그러나 인플레이션 전망은 국내외 경제 상황, 원자재 가격 변동, 환율 등 다양한 요인에 의해 영향을 받을 수 있으므로, 정부의 전망과 실제 인플레이션 경로 사이에는 차이가 발생할 수 있습니다.\n",
      "답변 :  미국의 현재 인플레이션 상황은 소비자 가격 지수(CPI)가 지난달 3.1% 상승하여 시장 예상치인 2.9%를 초과했습니다. 또한, 도매 가격을 반영하는 생산자 가격 지수(PPI)는 월간 0.3%, 연간 0.9% 증가하여 모두 시장 예측을 상회했습니다. 이는 강력한 고용 시장이 임금을 상승시키고, 결과적으로 서비스 가격을 끌어올리는 등의 요인으로 인해 인플레이션이 기대했던 것처럼 둔화되지 않고 있음을 나타냅니다. 이러한 상황은 지속적인 높은 인플레이션에 기여하고 있습니다.\n",
      "답변 :  중동 리스크의 지속은 올해 상반기에 인플레이션 둔화 추세가 상승세로 전환될 수 있는 불확실성에 기여하고 있습니다. 이는 부분적으로 러시아-우크라이나 갈등과 홍해 사건 이후 지난해 5월부터 상승 곡선을 그리고 있는 글로벌 공급망 압력 지수(GSCPI)가 글로벌 공급망 상황을 모니터링하기 때문입니다. GSCPI의 움직임은 글로벌 공급망 악화가 완화되고 있을 수는 있지만, 지속되는 지정학적 긴장이 미국의 인플레이션율에 상승 압력을 가할 수 있음을 시사합니다. GSCPI 변화와 인플레이션에 미치는 영향 사이에는 약 9개월의 시차가 있습니다. 따라서, 중동 리스크와 같은 지정학적 긴장은 인플레이션 상승 압력을 가할 수 있는 중요한 요인 중 하나로 간주됩니다.\n",
      "답변 :  한국 금리 정책의 현재 전망은 높은 금리가 장기화될 가능성을 시사하고 있습니다. 이러한 관점은 농산물 가격의 지속적인 상승과 국제유가의 반등 등 여러 경제적 요인에 의해 영향을 받으며, 이는 인플레이션 압력에 기여합니다. 또한, 미국의 강력한 고용 지표와 끈질기게 높은 서비스 가격을 특징으로 하는 글로벌 경제 환경은 상황을 더욱 복잡하게 만듭니다. 이러한 조건들은 미국 연방준비제도의 금리 인하 예상이 지연되어 달러가 강세를 보이는 결과를 초래했습니다. 따라서, 한국은 높은 금리, 높은 가격, 높은 환율의 도전에 직면해 있으며, 이러한 조건들이 장기간 지속될 가능성에 대한 우려가 있습니다.\n",
      "추론에 사용한 모델 : gpt-3.5-turbo-0125\n",
      "답변 :  과일 가격 상승은 전반적인 인플레이션에 상당한 영향을 미치며, 과일만으로도 전체 인플레이션율의 상당 부분을 차지합니다.\n",
      "답변 :  명절 이후 농산물 가격은 계속해서 상승하고 있습니다. 예를 들어, 귤의 가격은 지난 몇 년간의 평균 가격과 비교하여 두 배로 상승했습니다. 또한, 사과, 배, 시금치, 파, 오이 등 다른 품목들도 작년 대비 상당한 가격 상승을 보였습니다. 이러한 추세는 명절 이후 농산물의 가격이 계속해서 상승하고 있다는 것을 나타냅니다.\n",
      "답변 :  국제유가와 환율 상승은 물가 상승에 상당한 영향을 미치며, 전반적인 물가 상승에 기여합니다.\n",
      "답변 :  미국 연방준비제도의 금리 인하 시기가 미뤄진 이유는, 서비스 가격의 상당한 상승, 국제유가 상승, 물류 비용 증가로 인한 지속적인 높은 인플레이션 압력 때문입니다. 또한, 뉴욕 연방준비은행의 글로벌 공급망 압력 지수가 약간 상승했으며, 러시아-우크라이나 갈등 및 홍콩 상황과 같은 지정학적 리스크로 악화된 글로벌 공급망 도전에 대한 지속적인 증가를 보여줄 수 있습니다.\n",
      "답변 :  한국 내 주요 물가 변화에 대한 개요는 다음과 같습니다: 인플레이션이 높아져서 소비자물가가 상승하고, 과일가격도 상승하며 국제유가와 환율의 물가에 미치는 영향을 반영하고 있습니다. 이로 인해 물가가 상승에 대한 우려가 커지고 있으며, 서민들의 경제적 어려움이 심화되고 있습니다.\n",
      "답변 :  최근에는 휘발유 가격이 하락한 기간 이후 상승하고 있으며, 이는 국제유가의 반등과 미국 달러의 강세로 인한 것입니다.\n",
      "답변 :  정부는 인플레이션 둔화에 대한 전망에 대해 우려하고 있습니다.\n",
      "답변 :  미국의 현재 인플레이션 상황은 상승세를 보이고 있으며, 지난 달 소비자물가지수가 3.1% 상승하여 시장 기대를 넘어섰습니다. 또한, 도매물가를 나타내는 생산자물가지수도 전월 대비 0.3% 상승하고 작년 동월 대비 0.9% 상승하여 시장 예상을 초과했습니다. 고용 시장의 견조한 상황으로 인해 서비스 가격이 상승하고, 글로벌 공급망 압력 등이 인플레이션 압력에 기여하고 있습니다.\n",
      "답변 :  중동 리스크는 물가 상승에 기여하고 있습니다.\n",
      "답변 :  한국의 금리 정책의 현재 전망은 지속적으로 높은 인플레이션, 국제유가의 장기적인 상승, 그리고 대한민국 원화와 미국 달러 간의 동시 상승과 같은 요인들에 영향을 받고 있습니다. 이러한 요인들은 앞으로의 국가의 금리 정책 결정과정에 영향을 미칠 것으로 예상됩니다.\n",
      "10\n",
      "10\n"
     ]
    }
   ],
   "source": [
    "from llama_index import ServiceContext, VectorStoreIndex, SimpleDirectoryReader\n",
    "from llama_index.embeddings import OpenAIEmbedding\n",
    "from llama_index import ServiceContext\n",
    "from llama_index import set_global_service_context\n",
    "from llama_index.chat_engine.types import ChatMode\n",
    "from llama_index.llms import OpenAI\n",
    "\n",
    "#4turbo, 4, 3.5turbo\n",
    "testing_models = [\"gpt-4-0125-preview\",\"gpt-3.5-turbo-0125\"]\n",
    "embed_model = \"local:sentence-transformers/distiluse-base-multilingual-cased-v1\"\n",
    "\n",
    "gpt4t_ans = []\n",
    "gpt35t_ans = []\n",
    "\n",
    "for model in testing_models:\n",
    "    llm = OpenAI(model = model, temperature = 0.1)\n",
    "    service_context = ServiceContext.from_defaults(embed_model=embed_model,llm = llm, chunk_size=1024)\n",
    "    set_global_service_context(service_context)\n",
    "    documents = SimpleDirectoryReader(\"./news1/\").load_data()\n",
    "    index = VectorStoreIndex.from_documents(documents)\n",
    "    engine = index.as_chat_engine(response_mode=\"compact\",max_tokens=200)\n",
    "    print(f'추론에 사용한 모델 : {model}')\n",
    "    for question in questions_list:\n",
    "        message_input = (\n",
    "                    f\"Thinking through several stages, answer politely in Korean. {question}\"\n",
    "                )\n",
    "        res = engine.chat(message_input)\n",
    "        message = res.response\n",
    "        print(\"답변 : \", message)\n",
    "        if model == testing_models[0]:\n",
    "            gpt4t_ans.append(message)\n",
    "        else:\n",
    "            gpt35t_ans.append(message)\n",
    "\n",
    "print(len(gpt4t_ans))\n",
    "print(len(gpt35t_ans))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "45a00d40-70d4-4ce3-8a0d-c8ea8113f075",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>simple_Q</th>\n",
       "      <th>simple_Q_ans</th>\n",
       "      <th>GPT-4 Turbo</th>\n",
       "      <th>GPT-3.5 Turbo</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>과일 가격 상승이 인플레이션에 미치는 영향은?</td>\n",
       "      <td>과일 가격 상승이 전체 인플레이션의 약 7분의 1을 차지하여 중요한 요소임.</td>\n",
       "      <td>과일 가격 상승이 인플레이션에 미치는 영향은 다양합니다. 우선, 과일 가격 상승은 ...</td>\n",
       "      <td>과일 가격 상승은 전반적인 인플레이션에 상당한 영향을 미치며, 과일만으로도 전체 인...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>명절 이후 농산물 가격은 어떻게 변화하고 있는가?</td>\n",
       "      <td>명절 이후에도 농산물 가격이 계속해서 상승하고 있음.</td>\n",
       "      <td>명절 이후에도 농산물 가격은 계속 상승하고 있습니다. 명절 기간 동안 수요가 증가하...</td>\n",
       "      <td>명절 이후 농산물 가격은 계속해서 상승하고 있습니다. 예를 들어, 귤의 가격은 지난...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>국제유가와 환율 상승이 물가 상승에 미치는 영향은?</td>\n",
       "      <td>국제유가와 환율 상승으로 인해 물가상승률이 3%대에 이를 것으로 전망됨.</td>\n",
       "      <td>국제유가 상승과 환율 상승은 물가 상승에 다음과 같은 영향을 미칩니다.\\n\\n국제유...</td>\n",
       "      <td>국제유가와 환율 상승은 물가 상승에 상당한 영향을 미치며, 전반적인 물가 상승에 기...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>미국 금리 인하 시기가 미뤄진 이유는?</td>\n",
       "      <td>미국에서 서비스 물가 상승 지속으로 금리 인하 시기가 3월에서 6월로 미뤄짐.</td>\n",
       "      <td>미국 금리 인하 시기가 미뤄진 이유는 지속적인 높은 인플레이션, 강력한 고용 시장이...</td>\n",
       "      <td>미국 연방준비제도의 금리 인하 시기가 미뤄진 이유는, 서비스 가격의 상당한 상승, ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>한국 내 주요 물가 변화에 대한 개요는?</td>\n",
       "      <td>감귤, 사과, 배 등 과일 가격 크게 상승.</td>\n",
       "      <td>한국 내에서는 휘발유와 과일을 포함한 다양한 상품의 가격이 눈에 띄게 상승하고 있으...</td>\n",
       "      <td>한국 내 주요 물가 변화에 대한 개요는 다음과 같습니다: 인플레이션이 높아져서 소비...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>최근 휘발유 가격의 변화는?</td>\n",
       "      <td>국내 주유소의 휘발유·경유 가격이 3주 연속 상승.</td>\n",
       "      <td>최근 휘발유 가격은 3주 연속 상승세를 보이고 있습니다. 구체적으로, 2월 둘째 주...</td>\n",
       "      <td>최근에는 휘발유 가격이 하락한 기간 이후 상승하고 있으며, 이는 국제유가의 반등과 ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>인플레이션 둔화에 대한 정부의 전망은?</td>\n",
       "      <td>정부는 소비자물가 상승률이 이달과 다음 달에 3%대에 진입할 것으로 전망.</td>\n",
       "      <td>인플레이션 둔화에 대한 정부의 구체적인 전망은 명시되지 않았습니다. 그러나 일반적으...</td>\n",
       "      <td>정부는 인플레이션 둔화에 대한 전망에 대해 우려하고 있습니다.</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>미국 인플레이션 상황의 현재 상태는?</td>\n",
       "      <td>미국은 인플레이션 둔화의 마지막 단계에서 어려움을 겪으며, 소비자물가지수와 도매물가...</td>\n",
       "      <td>미국의 현재 인플레이션 상황은 소비자 가격 지수(CPI)가 지난달 3.1% 상승하여...</td>\n",
       "      <td>미국의 현재 인플레이션 상황은 상승세를 보이고 있으며, 지난 달 소비자물가지수가 3...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>중동 리스크가 물가 상승에 미치는 영향은?</td>\n",
       "      <td>중동 리스크 장기화가 인플레이션 상승 위험을 높일 수 있음.</td>\n",
       "      <td>중동 리스크의 지속은 올해 상반기에 인플레이션 둔화 추세가 상승세로 전환될 수 있는...</td>\n",
       "      <td>중동 리스크는 물가 상승에 기여하고 있습니다.</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>한국 금리 정책의 현재 전망은?</td>\n",
       "      <td>한국은행이 금리 인하 기대를 차단할 가능성이 커지며, 금리 인하 시점이 하반기로 밀...</td>\n",
       "      <td>한국 금리 정책의 현재 전망은 높은 금리가 장기화될 가능성을 시사하고 있습니다. 이...</td>\n",
       "      <td>한국의 금리 정책의 현재 전망은 지속적으로 높은 인플레이션, 국제유가의 장기적인 상...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                       simple_Q  \\\n",
       "0     과일 가격 상승이 인플레이션에 미치는 영향은?   \n",
       "1   명절 이후 농산물 가격은 어떻게 변화하고 있는가?   \n",
       "2  국제유가와 환율 상승이 물가 상승에 미치는 영향은?   \n",
       "3         미국 금리 인하 시기가 미뤄진 이유는?   \n",
       "4        한국 내 주요 물가 변화에 대한 개요는?   \n",
       "5               최근 휘발유 가격의 변화는?   \n",
       "6         인플레이션 둔화에 대한 정부의 전망은?   \n",
       "7          미국 인플레이션 상황의 현재 상태는?   \n",
       "8       중동 리스크가 물가 상승에 미치는 영향은?   \n",
       "9             한국 금리 정책의 현재 전망은?   \n",
       "\n",
       "                                        simple_Q_ans  \\\n",
       "0         과일 가격 상승이 전체 인플레이션의 약 7분의 1을 차지하여 중요한 요소임.   \n",
       "1                      명절 이후에도 농산물 가격이 계속해서 상승하고 있음.   \n",
       "2           국제유가와 환율 상승으로 인해 물가상승률이 3%대에 이를 것으로 전망됨.   \n",
       "3        미국에서 서비스 물가 상승 지속으로 금리 인하 시기가 3월에서 6월로 미뤄짐.   \n",
       "4                           감귤, 사과, 배 등 과일 가격 크게 상승.   \n",
       "5                       국내 주유소의 휘발유·경유 가격이 3주 연속 상승.   \n",
       "6          정부는 소비자물가 상승률이 이달과 다음 달에 3%대에 진입할 것으로 전망.   \n",
       "7  미국은 인플레이션 둔화의 마지막 단계에서 어려움을 겪으며, 소비자물가지수와 도매물가...   \n",
       "8                  중동 리스크 장기화가 인플레이션 상승 위험을 높일 수 있음.   \n",
       "9  한국은행이 금리 인하 기대를 차단할 가능성이 커지며, 금리 인하 시점이 하반기로 밀...   \n",
       "\n",
       "                                         GPT-4 Turbo  \\\n",
       "0  과일 가격 상승이 인플레이션에 미치는 영향은 다양합니다. 우선, 과일 가격 상승은 ...   \n",
       "1  명절 이후에도 농산물 가격은 계속 상승하고 있습니다. 명절 기간 동안 수요가 증가하...   \n",
       "2  국제유가 상승과 환율 상승은 물가 상승에 다음과 같은 영향을 미칩니다.\\n\\n국제유...   \n",
       "3  미국 금리 인하 시기가 미뤄진 이유는 지속적인 높은 인플레이션, 강력한 고용 시장이...   \n",
       "4  한국 내에서는 휘발유와 과일을 포함한 다양한 상품의 가격이 눈에 띄게 상승하고 있으...   \n",
       "5  최근 휘발유 가격은 3주 연속 상승세를 보이고 있습니다. 구체적으로, 2월 둘째 주...   \n",
       "6  인플레이션 둔화에 대한 정부의 구체적인 전망은 명시되지 않았습니다. 그러나 일반적으...   \n",
       "7  미국의 현재 인플레이션 상황은 소비자 가격 지수(CPI)가 지난달 3.1% 상승하여...   \n",
       "8  중동 리스크의 지속은 올해 상반기에 인플레이션 둔화 추세가 상승세로 전환될 수 있는...   \n",
       "9  한국 금리 정책의 현재 전망은 높은 금리가 장기화될 가능성을 시사하고 있습니다. 이...   \n",
       "\n",
       "                                       GPT-3.5 Turbo  \n",
       "0  과일 가격 상승은 전반적인 인플레이션에 상당한 영향을 미치며, 과일만으로도 전체 인...  \n",
       "1  명절 이후 농산물 가격은 계속해서 상승하고 있습니다. 예를 들어, 귤의 가격은 지난...  \n",
       "2  국제유가와 환율 상승은 물가 상승에 상당한 영향을 미치며, 전반적인 물가 상승에 기...  \n",
       "3  미국 연방준비제도의 금리 인하 시기가 미뤄진 이유는, 서비스 가격의 상당한 상승, ...  \n",
       "4  한국 내 주요 물가 변화에 대한 개요는 다음과 같습니다: 인플레이션이 높아져서 소비...  \n",
       "5  최근에는 휘발유 가격이 하락한 기간 이후 상승하고 있으며, 이는 국제유가의 반등과 ...  \n",
       "6                 정부는 인플레이션 둔화에 대한 전망에 대해 우려하고 있습니다.  \n",
       "7  미국의 현재 인플레이션 상황은 상승세를 보이고 있으며, 지난 달 소비자물가지수가 3...  \n",
       "8                          중동 리스크는 물가 상승에 기여하고 있습니다.  \n",
       "9  한국의 금리 정책의 현재 전망은 지속적으로 높은 인플레이션, 국제유가의 장기적인 상...  "
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import pandas as pd\n",
    "\n",
    "df = pd.DataFrame({\n",
    "    \"simple_Q\": simple_questions,\n",
    "    \"simple_Q_ans\": simple_answers,\n",
    "    \"GPT-4 Turbo\": gpt4t_ans,\n",
    "    \"GPT-3.5 Turbo\": gpt35t_ans\n",
    "})\n",
    "\n",
    "df.to_csv(\"simple_Q_gpt_models_comparison.csv\", index=False)\n",
    "\n",
    "df"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4ece9639-31e4-40dc-8fe9-61343dfbe535",
   "metadata": {},
   "source": [
    "# Advanced Q 답변 추론 /gpt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "275c9651-728e-41c5-9b59-5271683ba897",
   "metadata": {},
   "outputs": [],
   "source": [
    "questions_list = advanced_questions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "66a8367e-99a4-4352-9e7d-1d1166c5dc1d",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "추론에 사용한 모델 : gpt-4-0125-preview\n",
      "답변 :  한국 경제에 미치는 가계부채의 영향은 상당히 중대하며 여러 가지 측면이 있습니다. 우선, 가계부채가 약 1845조 원에 달해 GDP의 108%에 이르는 것은 가계가 부채를 감당하기 위해 수입이 부족한 상황에 처해 있음을 나타냅니다. 이러한 불균형은 많은 개인이 추가 부채 없이 재정적 의무를 관리할 수 없어 부채 상황을 악화시키는 대출의 순환을 초래합니다.\n",
      "\n",
      "둘째, 가계부채의 증가는 생활비와 대출 이자율이 상승하는 가운데 소득 수준이 정체된 상태에서 은행 대출을 상환할 수 없는 소규모 사업주와 자영업자의 증가와 연결되어 있습니다. 이 상황은 기록적인 채무 불이행과 금융 기관이 이러한 소규모 사업주를 대신해 대출을 처리하는 상황으로 이어져 이 부문 내에서 금융 불안정이 커지고 있음을 나타냅니다.\n",
      "\n",
      "게다가, 인천, 부산, 광주와 같은 도시에서 가계부채 증가의 지역적 차이는 다양한 지역에서 경제적 압박이 불균등하게 나타나고 있음을 강조합니다. 특히 인천은 임대 사기의 가장 높은 발생률과 심각한 역임대 상황으로 추가적인 도전에 직면하여 주민들의 재정적 안정성을 더욱 악화시킵니다.\n",
      "\n",
      "전반적인 경제적 영향은 부채 위기를 해결하기 위한 조치가 취해지지 않을 경우 증가된 사회적 위험과 더 깊은 경제 침체의 가능성에 의해 더욱 심화됩니다. 이 상황은 가계와 소규모 사업주의 부채 부담을 완화하기 위한 정부의 목표 지향적 개입이 필요함을 시사하며, 이러한 조치 없이는 한국의 경제적 및 사회적 구조가 심각한 압박을 받을 수 있음을 나타냅니다.\n",
      "답변 :  코로나19 이후 소상공인의 경제 상황은 악화되었습니다. 팬데믹이 종료되었음에도 불구하고 소상공인들은 코로나19 이전 수준으로 수입 수준이 회복되지 않았습니다. 또한, 수입 증가 없이 대출 만기일이 다가오는 도전에 직면하여, 팬데믹 기간보다 현재 상황이 더 어려워졌습니다. 기존 부채를 상환하기 위해 새로운 부채를 지는 소상공인의 수가 450만 명을 넘어서, 상당한 재정적 부담을 나타냅니다. 또한, 2020년 4월에 처음 도입된 대출 만기 연장 및 상환 유예와 같은 정부 지원 조치의 효과가 시간이 지남에 따라 감소하였으며, 다중 부채를 가진 소상공인의 총 대출 잔액이 사상 최고치에 도달하여 모든 소상공인 대출의 71.3%를 차지했습니다. 이로 인해 소상공인에 대한 재정적 부담이 증가하였으며, 대출 이자율이 0.25%포인트 증가할 경우 총 이자 및 개인별 평균 연간 이자가 상승할 것으로 추정됩니다.\n",
      "답변 :  부자증세는 정부에 추가적인 수입을 제공하여, 이를 소상공인 및 자영업자 지원에 할당할 수 있습니다. 이 추가 자금은 부채 상환에 어려움을 겪고 있는 소상공인을 위한 부채 탕감 프로그램을 시행하고 파산 면제 조치를 가속화하는 데 사용될 수 있습니다. 이러한 사업체에 대한 재정적 부담을 줄임으로써 정부는 경제를 안정시키고 광범위한 사업 실패와 관련된 사회적 위험을 완화할 수 있습니다. 또한, 이러한 접근 방식은 부의 더 공평한 분배에 기여하고 인구의 다양한 부문 간 경제적 격차를 줄일 수 있습니다.\n",
      "답변 :  은행 대출금리 상승은 소상공인에게 여러 가지 영향을 미칩니다. 이 중 하나는 은행 부채를 상환할 수 없는 소상공인의 수가 증가하는 것입니다. 이 상황은 경제 침체와 소비자 지출의 축소로 인해 악화되어, 소상공인 사이에서 채무 불이행이 크게 증가하였습니다. 결과적으로, 금융 기관과 신용보증기금이 이러한 소상공인을 대신해 대출 상환을 처리해야 했으며, 이는 이 부문 내에서 재정적 부담이 커지고 있음을 나타냅니다.\n",
      "답변 :  가계부채 증가가 개인과 지역 경제에 미치는 단기적 영향으로는 가계의 재정적 스트레스가 증가하여 소비 지출이 감소하는 것을 들 수 있습니다. 이러한 지출 감소는 기업의 수익 감소로 이어져 해고나 채용 축소로 이어질 수 있으며, 경제 성장이 둔화될 수 있습니다. 특히 인천과 같이 임대 사기와 역임대 상황이 만연한 지역의 지역 경제에는 가계부채 증가가 경제적 불균형을 심화시키고 지역 경제 위기의 위험을 깊게 할 수 있습니다.\n",
      "\n",
      "장기적으로는 가계부채의 증가가 주택 소유자와 소상공인 사이에서 채무 불이행과 파산의 증가로 이어질 수 있습니다. 이는 지역 경제에 연쇄적인 영향을 미쳐 실업률 증가와 전반적인 경제 안정성 감소로 이어질 수 있습니다. 증가된 재정적 불안정성은 지역에 대한 투자를 저해하여 경제 성장을 더욱 억제할 수 있습니다. 또한, 가계부채가 증가함에 따라 심각한 경제 침체의 가능성이 커지며, 이는 광범위한 사회적 및 경제적 도전을 포함하여 경제 침체를 심화시킬 수 있습니다.\n",
      "추론에 사용한 모델 : gpt-3.5-turbo-0125\n",
      "답변 :  가계부채가 한국 경제에 미치는 영향은 상당히 중요합니다. 가계부채의 증가로 인해 가계는 소득만으로 부채를 상환하기 어려운 상황에 이르고 있습니다. 이는 소상공인과 개인의 파산 증가로 이어져 심각한 경제 위험을 초래하고 있습니다. 또한, 증가하는 가계부채는 소비자 신뢰도의 하락과 경제 불안정성 감소에 기여하며, 경제 침체를 야기할 수 있습니다.\n",
      "답변 :  코로나19 이후 소상공인의 경제 상황은 더 어려워졌습니다. 소상공인들의 소득 수준은 코로나19 이전 수준으로 회복되지 않았고, 대출 만기일은 다가오는데 소득은 제자리인 상황입니다. 또한, 코로나19 시기보다 현재 상황이 소상공인들에게 더 어렵다고 할 수 있습니다.\n",
      "답변 :  부자증세가 정부 재정에 미치는 영향은 부자들로부터 세금을 추가로 걷어내어 정부의 수입을 증가시키는 것을 의미합니다. 이로 인해 정부는 추가적인 사회복지 및 경제지원 등을 할 수 있으며, 이는 경제적으로 부담을 받는 국민들에 대한 지원 및 경제 정책을 강화하는데 도움이 될 것입니다.\n",
      "\n",
      "부자증세로 인한 국내 소상공인의 경우 소득이 감소하고, 이로 인해 경영지원이 부족해지며, 부자들의 세금으로 인해 정부의 예산이 증가하면서 소상공인들에 대한 지원이 줄어들고, 부자들로부터 세금을 추가로 걷어내어 정부의 재정이 증가하면서 소상공인들에 대한 지원이 더 어려워질 수 있습니다.\n",
      "답변 :  은행 대출금리 상승은 소상공인에게 빚 상환 부담을 더욱 가중시킬 수 있습니다. 소상공인들이 대출금리 상승으로 인해 빚을 갚지 못하게 되면 '디폴트' 상황이 발생할 수 있으며, 이는 경제적 위험으로 이어질 수 있습니다.\n",
      "답변 :  가계부채 증가는 장기적으로는 가계의 경제적 불안정성을 증가시키고 개인의 소비 능력을 제약할 수 있으며, 지역 경제에는 부동산 시장 등에서의 불안정성을 초래할 수 있습니다. 단기적으로는 가계의 소비 감소로 인한 경기 위축과 소비자 심리 악화로 이어질 수 있습니다.\n",
      "5\n",
      "5\n"
     ]
    }
   ],
   "source": [
    "from llama_index import ServiceContext, VectorStoreIndex, SimpleDirectoryReader\n",
    "from llama_index.embeddings import OpenAIEmbedding\n",
    "from llama_index import ServiceContext\n",
    "from llama_index import set_global_service_context\n",
    "from llama_index.chat_engine.types import ChatMode\n",
    "from llama_index.llms import OpenAI\n",
    "\n",
    "testing_models = [\"gpt-4-0125-preview\",\"gpt-3.5-turbo-0125\"]\n",
    "embed_model = \"local:sentence-transformers/distiluse-base-multilingual-cased-v1\"\n",
    "\n",
    "gpt4t_ans = []\n",
    "gpt35t_ans = []\n",
    "\n",
    "for model in testing_models:\n",
    "    llm = OpenAI(model = model, temperature = 0.1)\n",
    "    service_context = ServiceContext.from_defaults(embed_model=embed_model,llm = llm, chunk_size=1024)\n",
    "    set_global_service_context(service_context)\n",
    "    documents = SimpleDirectoryReader(\"./news2/\").load_data()\n",
    "    index = VectorStoreIndex.from_documents(documents)\n",
    "    engine = index.as_chat_engine(response_mode=\"compact\",max_tokens=200)\n",
    "    print(f'추론에 사용한 모델 : {model}')\n",
    "    for question in questions_list:\n",
    "        message_input = (\n",
    "                    f\"Thinking through several stages, answer politely in Korean. {question}\"\n",
    "                )\n",
    "        res = engine.chat(message_input)\n",
    "        message = res.response\n",
    "        print(\"답변 : \", message)\n",
    "        if model == testing_models[0]:\n",
    "            gpt4t_ans.append(message)\n",
    "        else:\n",
    "            gpt35t_ans.append(message)\n",
    "\n",
    "print(len(gpt4t_ans))\n",
    "print(len(gpt35t_ans))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "e722c716-2f4b-4b7d-80e0-52686372c027",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>advanced_Q</th>\n",
       "      <th>advanced_Q_ans</th>\n",
       "      <th>GPT-4 Turbo</th>\n",
       "      <th>GPT-3.5 Turbo</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>가계부채가 한국 경제에 미치는 영향에 대해 설명하세요.</td>\n",
       "      <td>가계부채 증가는 개인의 소비를 억제하고, 금리 상승 시 이자 부담 증가로 이어져 가...</td>\n",
       "      <td>한국 경제에 미치는 가계부채의 영향은 상당히 중대하며 여러 가지 측면이 있습니다. ...</td>\n",
       "      <td>가계부채가 한국 경제에 미치는 영향은 상당히 중요합니다. 가계부채의 증가로 인해 가...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>코로나19 이후 소상공인의 경제 상황 변화는 어떠한가요?</td>\n",
       "      <td>코로나19 이후 소상공인은 매출 감소와 빚 증가로 인해 경제적 어려움을 겪고 있으며...</td>\n",
       "      <td>코로나19 이후 소상공인의 경제 상황은 악화되었습니다. 팬데믹이 종료되었음에도 불구...</td>\n",
       "      <td>코로나19 이후 소상공인의 경제 상황은 더 어려워졌습니다. 소상공인들의 소득 수준은...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>부자증세가 정부 재정과 소상공인 지원에 어떻게 영향을 미칠까요?</td>\n",
       "      <td>부자증세를 통한 추가 재정 확보는 정부와 지자체가 소상공인 지원과 같은 사회 안전망...</td>\n",
       "      <td>부자증세는 정부에 추가적인 수입을 제공하여, 이를 소상공인 및 자영업자 지원에 할당...</td>\n",
       "      <td>부자증세가 정부 재정에 미치는 영향은 부자들로부터 세금을 추가로 걷어내어 정부의 수...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>은행 대출금리 상승이 소상공인에게 미치는 영향은 무엇인가요?</td>\n",
       "      <td>은행 대출금리 상승은 소상공인의 대출 이자 부담을 증가시켜 재정적 어려움을 가중시키...</td>\n",
       "      <td>은행 대출금리 상승은 소상공인에게 여러 가지 영향을 미칩니다. 이 중 하나는 은행 ...</td>\n",
       "      <td>은행 대출금리 상승은 소상공인에게 빚 상환 부담을 더욱 가중시킬 수 있습니다. 소상...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>가계부채 증가가 개인과 지역 경제에 끼치는 장단기적 영향은 무엇인가요?</td>\n",
       "      <td>가계부채 증가는 단기적으로는 생활비 마련을 위한 대출 의존도를 높이지만, 장기적으로...</td>\n",
       "      <td>가계부채 증가가 개인과 지역 경제에 미치는 단기적 영향으로는 가계의 재정적 스트레스...</td>\n",
       "      <td>가계부채 증가는 장기적으로는 가계의 경제적 불안정성을 증가시키고 개인의 소비 능력을...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                advanced_Q  \\\n",
       "0           가계부채가 한국 경제에 미치는 영향에 대해 설명하세요.   \n",
       "1          코로나19 이후 소상공인의 경제 상황 변화는 어떠한가요?   \n",
       "2      부자증세가 정부 재정과 소상공인 지원에 어떻게 영향을 미칠까요?   \n",
       "3        은행 대출금리 상승이 소상공인에게 미치는 영향은 무엇인가요?   \n",
       "4  가계부채 증가가 개인과 지역 경제에 끼치는 장단기적 영향은 무엇인가요?   \n",
       "\n",
       "                                      advanced_Q_ans  \\\n",
       "0  가계부채 증가는 개인의 소비를 억제하고, 금리 상승 시 이자 부담 증가로 이어져 가...   \n",
       "1  코로나19 이후 소상공인은 매출 감소와 빚 증가로 인해 경제적 어려움을 겪고 있으며...   \n",
       "2  부자증세를 통한 추가 재정 확보는 정부와 지자체가 소상공인 지원과 같은 사회 안전망...   \n",
       "3  은행 대출금리 상승은 소상공인의 대출 이자 부담을 증가시켜 재정적 어려움을 가중시키...   \n",
       "4  가계부채 증가는 단기적으로는 생활비 마련을 위한 대출 의존도를 높이지만, 장기적으로...   \n",
       "\n",
       "                                         GPT-4 Turbo  \\\n",
       "0  한국 경제에 미치는 가계부채의 영향은 상당히 중대하며 여러 가지 측면이 있습니다. ...   \n",
       "1  코로나19 이후 소상공인의 경제 상황은 악화되었습니다. 팬데믹이 종료되었음에도 불구...   \n",
       "2  부자증세는 정부에 추가적인 수입을 제공하여, 이를 소상공인 및 자영업자 지원에 할당...   \n",
       "3  은행 대출금리 상승은 소상공인에게 여러 가지 영향을 미칩니다. 이 중 하나는 은행 ...   \n",
       "4  가계부채 증가가 개인과 지역 경제에 미치는 단기적 영향으로는 가계의 재정적 스트레스...   \n",
       "\n",
       "                                       GPT-3.5 Turbo  \n",
       "0  가계부채가 한국 경제에 미치는 영향은 상당히 중요합니다. 가계부채의 증가로 인해 가...  \n",
       "1  코로나19 이후 소상공인의 경제 상황은 더 어려워졌습니다. 소상공인들의 소득 수준은...  \n",
       "2  부자증세가 정부 재정에 미치는 영향은 부자들로부터 세금을 추가로 걷어내어 정부의 수...  \n",
       "3  은행 대출금리 상승은 소상공인에게 빚 상환 부담을 더욱 가중시킬 수 있습니다. 소상...  \n",
       "4  가계부채 증가는 장기적으로는 가계의 경제적 불안정성을 증가시키고 개인의 소비 능력을...  "
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df = pd.DataFrame({\n",
    "    \"advanced_Q\": advanced_questions,\n",
    "    \"advanced_Q_ans\": advanced_answers,\n",
    "    \"GPT-4 Turbo\": gpt4t_ans,\n",
    "    \"GPT-3.5 Turbo\": gpt35t_ans\n",
    "})\n",
    "\n",
    "df.to_csv(\"advanced_Q_gpt_models_comparison.csv\", index=False)\n",
    "\n",
    "df"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d94ebcd4-3e6a-4b1f-8ab7-81e6f286288b",
   "metadata": {},
   "source": [
    "# LDCC/LDCC-SOLAR-10.7B, mncai/llama2-13b-dpo-v7"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "55fd6150-0299-4f23-97e0-24b4a1d96eb8",
   "metadata": {},
   "source": [
    "## ldcc,llama / simple Q 답변 추론 /local"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "64abf47c-e890-4566-842f-f453636171a1",
   "metadata": {},
   "outputs": [],
   "source": [
    "questions_list = simple_questions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "id": "1ad7d0b0-ddd3-4aae-b437-562108b0232e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "db18419123aa47c98d72f171a3c12c2a",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/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"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "답변 :  과일 가격 이 인플레이 션에 미치는  영향은 과일이 소비자들의  식료품 지출에 큰 비중을  차지하기  때문에  소비 자물가 상승에 큰 요인으로  작용 할 수 있다 . 통계청에  따르면  지난 달 소비 자물가지 수 상승 에 과실 부문 이 0.4%포인트 의 기여도를  보였는데, 이는  2011년 1월(0.4%포인트 ) 이후  최대치였다 . 과일 가격 이 계속 오르 면 전체  인플레이 션 상승 에 기여할 가능성이  높 다. 따라 서 과일 가격 의 안정 을 위한  정책 적 노력이  필요할 것으로  보인다 . The 2018–19 season was the 118th season in the existence of Real Madrid and the club's 87th consecutive season in the top flight of Spanish\n",
      "답변 :  명절 이 지나 도 농산물  가격 은 여전히  고공행진 중이다 . 통계청에  따르면  지난  1월 소비 자물가 상승률 의 기여도 1위를  차지한 것은 과일(0.4%포인트 )이었다 . 과일 가격 은 지난 해 12월 대비  17.4% 올랐다 . 귤 은 평년 대비  두배 이상 에 달하는  1만 원대까지  오르기도  했다 . 채소도 1월 소비 자물가 상승률 의 기여도가  0.2%포인트 에 달했다 . 시금치 한 단은 4000원, 애호박 한 개는 3000원, 대파 한 봉 은 5000원이 다.\n",
      "\n",
      "이처럼  농산물 이 물가 상승 을 주도하는  데 국제유가 와 원·달러  환율 의 동반 \n",
      "답변 :  국제유가 와 환율  상승 은 수입 물 가격 이 오르 게 돼  소비 자물가에  반영되는  요인이  된다 . 한국은행 에 따르면  국제유가 와 원·달러  환율  상승 의 영향으로  지난 달 수입 물가지수는  전월 대비  2.2% 올라 3개월  만에  상승  전환했다 . 수입 물가지수는  1~3개월 의 시차를  두고  소비 자물가에  반영된다 .\n",
      "\n",
      "이러한 상황에서  미국 의 인플레이 션 둔화에  대한  기대감이  약해지면서  미국  중앙은행(Fed)이 금리  인하 시기 를 연기할 것이라는  분위기 가 형성되고  있다 . 미국  노동부가  발표 한 지난 달 소비 자 물가지 수(CPI)는 1년 전보다  6.5% 올라 11월(7.3%)과 12월(\n",
      "답변 :  미국  중앙은행 (Fed)은 지난 달 금리 를 0.25%포인트   올린 데 이어  이달에도  0.25%포인트   추가  인상을  결정했다 . 하지만  이달 회의에서  금리  인하 시기 에 대한  언급을 빼 고 대신  ‘인플레 가속화 추세 가 뚜렷하다 ’고 강조한 데 따라 금리  인하 시기 가 미뤄 질 것이라는  전망이  나왔다 . 이는  미국 의 서비스 업 부문 의 인플레이 션이 예상보다  높 은 것으로  나타났 기 때문이 다. 미국  노동부가  발표 한 지난 달 서비스 업 생산 자 물가지수는  1년 전보다  6.2% 올라 시장 예상치(5.4%)를 뛰 어넘었다 . 서비스 업은 미국 의 경제 에서 가장  큰 비중을\n",
      "답변 :  한국 내 주요 물가 변화 에 대한  개요는 다음과  같습니다 .\n",
      "\n",
      "1. 국제 유가 가 반등하며  국내  주유소 휘 발유·경유 주간  평균  판매  가격 이 3주째 올랐다 . 17일 한국 석유공사  유가정보시스템  오피넷 에 따르면  2월 둘 째 주(11∼15일) 전국  주유소 휘 발유 평균  판매가는  직전 주보다  L당 13.2원 오른  1천609.5원이었다 .\n",
      "\n",
      "2. 국제 유가 와 원·달러  환율 의 동반 상승 은 둔 화하던  인플레이션을  다시  자극하는  요인이다 . 달러 인덱스는  지난 해 말 100선까지  하락 한 뒤  이달 중순 104선까지  오르 며 원·달러  환율은  지난 해 말 \n",
      "답변 :  최근  휘 발유 가격 은 하락하던  추세에서  다시  상승 세로  돌아섰다 . 한국 석유공사  오피넷 에 따르면  전국  주유소 평균  휘 발유 판매가격 은 지난 달 넷 째주 리터당 1563.7원에서  이달 둘 째주 1609.5원으로  45.8원(2.9%) 올랐다 . 휘 발유 판매  가격 은 지난 달 넷 째주까지  16주 연속  하락 한 뒤  3주 연속  상승했다 . 휘 발유 가격 은 단기간  내에  국제 유가 와 원·달러  환율 의 변동에  반영될 것으로  보인다 .\n",
      "\n",
      "국제 유가 와 원·달러  환율 의 동반 상승 은 둔 화하던  인플레이션을  다시  자극하는  요인이다 . 달러 인덱스는  지난 해 말 10\n",
      "답변 :  기사 에 따르면  국제유가 와 원·달러  환율 의 동반 상승 에 “이달 물가상승률  3%대” 전망 이 나왔다 . 하지만  미국 의 경우 서비스 물가 상승 이 이어지면서  금리 인하 시기 가 3월에서  6월로 늦춰진 것으로  알려졌다 . 따라 서 국내에서도  인플레이 션 둔화에  대한  정부의  전망 이 달라질 가능성이  있다 . 하지만  정부는  인플레이 션 둔화에  대한  전망 을 밝힌  바 있으므로  정확한 상황 은 정부 의 공식  발표 를 기다려봐야  할 것입니다 .\n",
      "답변 :  미국  인플레이션은  2022년 6월 이후  둔화되고  있는  추세이다 . 미국  노동통계 국(BLS)에 따르면  지난 달 미국  소비 자물가 상승률은  6.5%로 6개월 째 연속  둔 화했다 . 하지만  이는  여전히  높 은 수준이며 , 연방준비제도이사 회(Fed)의 2% 인플레이 션 타깃 을 크게 웃돈 다. 전문가들은  미국  인플레이 션이 여전히  높 은 수준이고 , 연준이 추가  금리  인상을  통해 인플레이션을  잡 아내겠다는  의지를  강조한 다며 금리  인상 여지가  여전하다고  보고  있다 . 따라 서 미국  인플레이 션이 완전히  잡힐  때까지  고금리  기조가  이어질 가능성이  있다 . 하지만  최근 의 인플레이 션 둔화 추세 는 긍 정적인\n",
      "답변 :  기사 에 따르면 서 중동 리스크가  물가 상승 에 미치는  영향에 대한  정확한 답은 알 수 없습니다 . 기사 는 국제 유가 와 원·달러  환율 의 동반 상승 이 인플레이션을  다시  자극하고  있다 고 언급하고  있으나 , 이러한 상승 이 중동 리스크로  인한  것인지 는 명시되어  있지 않습니다 .\n",
      "답변 :  기사 에 따르면  \"미국  ‘끈적한  서비스  물가’에 금리  인하 3월 → 6월로\"라는  문구가  있는 것을  볼  때, 한국 의 금리 인하 전망 은 3월이 아닌  6월로 늦춰 질 가능성이  있는 것으로  보인다 . 그러나  이는  기사 에 따른 전망 일 뿐이며 , 한국은행 의 공식  입장은 아닌  것으로  보인다 . 따라 서 궁 금한 점이  더 있다 면 한국은행 의 공식  발표 나 자료를  참고하는  것이  좋을 것입니다 .\n",
      "10\n"
     ]
    }
   ],
   "source": [
    "import torch\n",
    "from transformers import BitsAndBytesConfig\n",
    "from llama_index.prompts import PromptTemplate\n",
    "from llama_index.llms import HuggingFaceLLM\n",
    "from llama_index import ServiceContext, VectorStoreIndex, SimpleDirectoryReader\n",
    "from llama_index.embeddings import OpenAIEmbedding\n",
    "from llama_index import ServiceContext\n",
    "from llama_index import set_global_service_context\n",
    "from llama_index.chat_engine.types import ChatMode\n",
    "from llama_index.llms import OpenAI\n",
    "\n",
    "device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
    "#\"LDCC/LDCC-SOLAR-10.7B\",mncai/llama2-13b-dpo-v7\n",
    "\n",
    "\n",
    "bnb_config = BitsAndBytesConfig(  \n",
    "    load_in_4bit= True,\n",
    "    bnb_4bit_quant_type= \"nf4\",\n",
    "    bnb_4bit_compute_dtype= torch.bfloat16,\n",
    "    bnb_4bit_use_double_quant= False,\n",
    ")\n",
    "\n",
    "testing_models = \"LDCC/LDCC-SOLAR-10.7B\"\n",
    "embed_model = \"local:sentence-transformers/distiluse-base-multilingual-cased-v1\"\n",
    "\n",
    "ldcc_ans = []\n",
    "\n",
    "llm = HuggingFaceLLM(\n",
    "    model_name = testing_models,\n",
    "    tokenizer_name= testing_models,\n",
    "    query_wrapper_prompt=PromptTemplate(\"<s> [INST] {query_str} [/INST] \"),\n",
    "    model_kwargs={\"quantization_config\": bnb_config},\n",
    "    generate_kwargs={\"temperature\": 0.1,},\n",
    "    device_map=\"auto\",\n",
    "    )\n",
    "service_context = ServiceContext.from_defaults(embed_model=embed_model,llm = llm,chunk_size=1024)\n",
    "set_global_service_context(service_context)\n",
    "documents = SimpleDirectoryReader(\"./news1/\").load_data()\n",
    "index = VectorStoreIndex.from_documents(documents)\n",
    "engine = index.as_query_engine(response_mode=\"compact\",max_tokens=200)\n",
    "for question in questions_list:\n",
    "    message_input = (\n",
    "                f\"Thinking through several stages, answer politely in Korean. {question}\"\n",
    "            )\n",
    "    res = engine.query(message_input)\n",
    "    message = res.response\n",
    "    print(\"답변 : \", message)\n",
    "    ldcc_ans.append(message)\n",
    "\n",
    "print(len(ldcc_ans))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "id": "71768b83-60c6-4717-a44c-f94f9d60bfab",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "68e84a4f178c4c9887a2cf8a623bd3d7",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Loading checkpoint shards:   0%|          | 0/3 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "답변 :  귤값 2배 뛰고 휘발유값 꿈틀 ... 발목 잡힌 인플레 둔화, 고금리 길어지나.txt\n",
      "\n",
      "귤값 2배 뛰고 휘발유값 꿈틀 ... 발목 잡힌 인플레 둔화, 고금리 길어지나\n",
      "명절 지나도 농산물 가격 고공행진 ... 휘발유 가격도 상승 반전\n",
      "국제유가·환율 동반 상승에 “이달 물가상승률 3%대” 전망\n",
      "美 ‘끈적한 서비스 물가’에 금리 인하 3월 → 6월로\n",
      "고금리·고물가·고환율 ‘3고’ 장기화하나\n",
      "\n",
      "과일 가격 상승은 인플레이션에 여러 가지 영향을 미칩니다. 첫째, 과일 가격 상승은 식료품 가격 상승으로 이어져 소비자 물가 상승에 기여합니다. 둘째, 과일 가격 상승은 농산물 가격 상승으로 이어져 전반적인 식품 가격 상승으로 이어집니다. 셋째, 과일 가격 상승은 원재료 가격 상승으로 이어져 다른 제품의 가격 상승으로 이어집니다. 넷째, 과일 가격 상승은 원화 약세로 이어져 수입품 가격 상승으로 이어져 소비자 물가 상승에 기여합니다. 다섯째, 과일 가격 상승은 전반적인 물가 상승으로 이어져\n",
      "답변 :  귤값 2배 뛰고 휘발유값 꿈틀 ... 발목 잡힌 인플레 둔화, 고금리 길어지나.txt\n",
      "\n",
      "명절 이후 농산물 가격은 상승세를 이어가고 있습니다. 통계청 국가통계포털(KOSIS) 자료에 따르면 지난 1월 소비자물가에서 '과실'의 기여도는 0.4%포인트로, 2011년 1월(0.4%p) 이후로 13년 만에 최대치를 기록했습니다. 1월 물가상승률(2.8%) 가운데 과일만으로 전체 인플레이션의 7분의 1을 끌어올렸습니다.\n",
      "답변 :  1. 국제유가와 환율 상승은 수입 물가에 영향을 미쳐 국내 소비자 물가에 반영됩니다. 2. 수입 물가 상승은 국내 생산자 및 소비자 물가 상승으로 이어집니다. 3. 국제유가와 환율 상승은 수입 물가 상승으로 이어져 국내 소비자 물가 상승을 더욱 가속화합니다. 4. 이러한 인플레이션 압력은 중앙은행이 인플레이션을 억제하기 위해 금리를 인상하는 요인이 됩니다.\n",
      "답변 :  1. 국제유가와 원·달러 환율의 동반 상승으로 인플레이션이 다시 상승할 수 있다는 우려 2. 1월 소비자물가에서 과일 기여도가 0.4%p로 13년 만에 최대치를 기록하는 등 농산물 가격의 고공행진 3. 2월 소비자물가 상승률이 3%대로 예상되는 등 인플레이션 둔화 속도가 느려질 수 있다는 우려 4. 美 서비스 물가 상승률이 6.2%로 40년 만에 최고치를 기록하는 등 서비스 물가의 지속적인 상승 5. 이러한 요인으로 인해 미국 연방준비제도(Fed)가 인플레이션을 억제하기 위해 금리를 인하하는 시기가 3월에서 6월로 미뤄졌습니다.\n",
      "답변 :  귤값 2배 뛰고 휘발유값 꿈틀 ... 발목 잡힌 인플레 둔화, 고금리 길어지나.txt\n",
      "\n",
      "한국 내 주요 물가 변화에 대한 개요는 다음과 같습니다:\n",
      "\n",
      "1. 농산물 가격 상승: 명절 이후에도 농산물 가격이 고공행진하고 있습니다.\n",
      "2. 휘발유 가격 상승: 국제유가와 원·달러 환율의 동반 상승으로 인해 휘발유 가격도 상승세로 돌아섰습니다.\n",
      "3. 인플레이션 둔화: 그러나 국제유가와 원·달러 환율의 상승으로 인해 인플레이션이 둔화되는 속도가 느려지고 있습니다.\n",
      "4. 금리 인상 전망: 이러한 인플레이션 둔화로 인해 미국 연방준비제도(Fed)는 금리 인상 속도를 늦출 것으로 예상됩니다.\n",
      "5. 고금리, 고물가, 고환율의 장기화 가능성: 인플레이션 둔화가 느려지고 금리 인상 속도가 늦춰지면서 고금리, 고물가, 고환율의 장기화가 우려되고 있습니다.\n",
      "답변 :  17일 한국석유공사 유가정보시스템 오피넷에 따르면 2월 둘째 주(11∼15일) 전국 주유소 휘발유 평균 판매가는 직전 주보다 L당 13.2원 오른 1천609.5원이었습니다. 휘발유 가격은 지난달 넷째주까지 16주 연속 하락했지만 3주 연속 상승했습니다.\n",
      "답변 :  1. 인플레이션 둔화에 대한 정부의 전망은 현재 인플레이션이 둔화되고 있지만, 여전히 높은 수준을 유지하고 있기 때문에 당분간 높은 인플레이션이 지속될 것으로 예상하고 있습니다. 정부는 인플레이션을 완화하기 위해 금리 인상과 재정 건전성 유지 등 다양한 정책을 시행하고 있습니다. 그러나 우크라이나 전쟁과 공급망 차질, 에너지 가격 상승 등 인플레이션을 자극하는 요인이 여전히 남아있어 인플레이션이 완전히 완화되기까지는 시간이 걸릴 것으로 예상하고 있습니다.\n",
      "답변 :  美 인플레이션 상황의 현재 상태는 둔화 속도가 느려지고 있으며, 국제유가와 원·달러 환율의 동반 상승으로 인해 3%대의 인플레이션이 예상되는 등 인플레이션이 다시 상승할 가능성이 있습니다.\n",
      "답변 :  1단계: 중동 리스크의 정의 이해\n",
      "중동 리스크는 중동 지역의 지정학적 불안정, 정치적 분쟁, 테러리즘, 석유 공급에 대한 위협 등 중동 지역에서 발생하는 사건이나 사건이 경제에 미치는 영향을 의미합니다.\n",
      "\n",
      "2단계: 중동 리스크가 물가 상승에 미치는 영향 분석\n",
      "중동 리스크는 석유 공급에 대한 위협으로 인해 석유 가격 상승으로 이어질 수 있습니다. 석유는 다양한 산업의 원료이자 운송 연료이므로 석유 가격이 상승하면 다른 상품의 가격도 상승할 수 있습니다. 또한 중동 지역의 불안정성은 투자자들의 투자 심리를 위축시켜 금융 시장에 영향을 미칠 수 있으며, 이는 다시 물가 상승으로 이어질 수 있습니다.\n",
      "\n",
      "3단계: 중동 리스크가 물가 상승에 미치는 영향의 예시\n",
      "중동 리스크가 물가 상승에 미치는 영향은 2020년 코로나19 팬데믹 당시에 잘 나타났습니다. 당시 중동 지역의 불안정성과 석유 공급에 대한 위협으로 인해 석유 가격이 상승했고, 이는 다른 상품의 가격 상승으로 이어졌습니다. 또한 금융 시장의 불안정으로 인해 투자 심리가 위축되어 주식과 채권 가격이 하락했고, 이는 다시 물가 상승으로 이어졌습니다.\n",
      "\n",
      "4단계: 중동 리스크가 물가 상승에 미치는 영향의 장기적 영향\n",
      "중동 리스크가 물가 상승에 미치는 영향은 단기적으로 발생할\n",
      "답변 :  \n",
      "한국의 금리 정책 전망은 현재 인플레이션이 둔화되고 있지만 여전히 높은 수준을 유지하고 있기 때문에 당분간 고금리 정책이 유지될 것으로 예상됩니다. 한국은행의 목표는 물가 안정이기 때문에 인플레이션이 목표치인 2%로 안정될 때까지 금리 인상을 계속할 것으로 예상됩니다. 그러나 국제 유가와 환율의 변동성, 우크라이나 전쟁과 같은 지정학적 리스크, 글로벌 경기 둔화 등 다양한 요인이 금리 결정에 영향을 미칠 수 있습니다. 따라서 금리 정책은 이러한 요인에 따라 변경될 수 있습니다.\n",
      "10\n"
     ]
    }
   ],
   "source": [
    "testing_models = \"mncai/llama2-13b-dpo-v7\"\n",
    "embed_model = \"local:sentence-transformers/distiluse-base-multilingual-cased-v1\"\n",
    "\n",
    "llama_ans = []\n",
    "\n",
    "llm = HuggingFaceLLM(\n",
    "    model_name = testing_models,\n",
    "    tokenizer_name= testing_models,\n",
    "    query_wrapper_prompt=PromptTemplate(\"<s> [INST] {query_str} [/INST] \"),\n",
    "    model_kwargs={\"quantization_config\": bnb_config},\n",
    "    generate_kwargs={\"temperature\": 0.1,},\n",
    "    device_map=\"auto\",\n",
    "    )\n",
    "service_context = ServiceContext.from_defaults(embed_model=embed_model,llm = llm,chunk_size=1024)\n",
    "set_global_service_context(service_context)\n",
    "documents = SimpleDirectoryReader(\"./news1/\").load_data()\n",
    "index = VectorStoreIndex.from_documents(documents)\n",
    "engine = index.as_query_engine(response_mode=\"compact\",max_tokens=200)\n",
    "for question in questions_list:\n",
    "    message_input = (\n",
    "                f\"Thinking through several stages, answer politely in Korean. {question}\"\n",
    "            )\n",
    "    res = engine.query(message_input)\n",
    "    message = res.response\n",
    "    print(\"답변 : \", message)\n",
    "    llama_ans.append(message)\n",
    "\n",
    "print(len(llama_ans))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "id": "2398fdc2-5dc7-4c4b-9ec7-fa2c677fad38",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>simple_Q</th>\n",
       "      <th>simple_Q_ans</th>\n",
       "      <th>ldcc_ans</th>\n",
       "      <th>llama_ans</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>과일 가격 상승이 인플레이션에 미치는 영향은?</td>\n",
       "      <td>과일 가격 상승이 전체 인플레이션의 약 7분의 1을 차지하여 중요한 요소임.</td>\n",
       "      <td>과일 가격 이 인플레이 션에 미치는  영향은 과일이 소비자들의  식료품 지출에 큰 ...</td>\n",
       "      <td>귤값 2배 뛰고 휘발유값 꿈틀 ... 발목 잡힌 인플레 둔화, 고금리 길어지나.tx...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>명절 이후 농산물 가격은 어떻게 변화하고 있는가?</td>\n",
       "      <td>명절 이후에도 농산물 가격이 계속해서 상승하고 있음.</td>\n",
       "      <td>명절 이 지나 도 농산물  가격 은 여전히  고공행진 중이다 . 통계청에  따르면 ...</td>\n",
       "      <td>귤값 2배 뛰고 휘발유값 꿈틀 ... 발목 잡힌 인플레 둔화, 고금리 길어지나.tx...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>국제유가와 환율 상승이 물가 상승에 미치는 영향은?</td>\n",
       "      <td>국제유가와 환율 상승으로 인해 물가상승률이 3%대에 이를 것으로 전망됨.</td>\n",
       "      <td>국제유가 와 환율  상승 은 수입 물 가격 이 오르 게 돼  소비 자물가에  반영되...</td>\n",
       "      <td>1. 국제유가와 환율 상승은 수입 물가에 영향을 미쳐 국내 소비자 물가에 반영됩니다...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>미국 금리 인하 시기가 미뤄진 이유는?</td>\n",
       "      <td>미국에서 서비스 물가 상승 지속으로 금리 인하 시기가 3월에서 6월로 미뤄짐.</td>\n",
       "      <td>미국  중앙은행 (Fed)은 지난 달 금리 를 0.25%포인트   올린 데 이어  ...</td>\n",
       "      <td>1. 국제유가와 원·달러 환율의 동반 상승으로 인플레이션이 다시 상승할 수 있다는 ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>한국 내 주요 물가 변화에 대한 개요는?</td>\n",
       "      <td>감귤, 사과, 배 등 과일 가격 크게 상승.</td>\n",
       "      <td>한국 내 주요 물가 변화 에 대한  개요는 다음과  같습니다 .\\n\\n1. 국제 유...</td>\n",
       "      <td>귤값 2배 뛰고 휘발유값 꿈틀 ... 발목 잡힌 인플레 둔화, 고금리 길어지나.tx...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>최근 휘발유 가격의 변화는?</td>\n",
       "      <td>국내 주유소의 휘발유·경유 가격이 3주 연속 상승.</td>\n",
       "      <td>최근  휘 발유 가격 은 하락하던  추세에서  다시  상승 세로  돌아섰다 . 한국...</td>\n",
       "      <td>17일 한국석유공사 유가정보시스템 오피넷에 따르면 2월 둘째 주(11∼15일) 전국...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>인플레이션 둔화에 대한 정부의 전망은?</td>\n",
       "      <td>정부는 소비자물가 상승률이 이달과 다음 달에 3%대에 진입할 것으로 전망.</td>\n",
       "      <td>기사 에 따르면  국제유가 와 원·달러  환율 의 동반 상승 에 “이달 물가상승률 ...</td>\n",
       "      <td>1. 인플레이션 둔화에 대한 정부의 전망은 현재 인플레이션이 둔화되고 있지만, 여전...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>미국 인플레이션 상황의 현재 상태는?</td>\n",
       "      <td>미국은 인플레이션 둔화의 마지막 단계에서 어려움을 겪으며, 소비자물가지수와 도매물가...</td>\n",
       "      <td>미국  인플레이션은  2022년 6월 이후  둔화되고  있는  추세이다 . 미국  ...</td>\n",
       "      <td>美 인플레이션 상황의 현재 상태는 둔화 속도가 느려지고 있으며, 국제유가와 원·달러...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>중동 리스크가 물가 상승에 미치는 영향은?</td>\n",
       "      <td>중동 리스크 장기화가 인플레이션 상승 위험을 높일 수 있음.</td>\n",
       "      <td>기사 에 따르면 서 중동 리스크가  물가 상승 에 미치는  영향에 대한  정확한 답...</td>\n",
       "      <td>1단계: 중동 리스크의 정의 이해\\n중동 리스크는 중동 지역의 지정학적 불안정, 정...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>한국 금리 정책의 현재 전망은?</td>\n",
       "      <td>한국은행이 금리 인하 기대를 차단할 가능성이 커지며, 금리 인하 시점이 하반기로 밀...</td>\n",
       "      <td>기사 에 따르면  \"미국  ‘끈적한  서비스  물가’에 금리  인하 3월 → 6월로...</td>\n",
       "      <td>\\n한국의 금리 정책 전망은 현재 인플레이션이 둔화되고 있지만 여전히 높은 수준을 ...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                       simple_Q  \\\n",
       "0     과일 가격 상승이 인플레이션에 미치는 영향은?   \n",
       "1   명절 이후 농산물 가격은 어떻게 변화하고 있는가?   \n",
       "2  국제유가와 환율 상승이 물가 상승에 미치는 영향은?   \n",
       "3         미국 금리 인하 시기가 미뤄진 이유는?   \n",
       "4        한국 내 주요 물가 변화에 대한 개요는?   \n",
       "5               최근 휘발유 가격의 변화는?   \n",
       "6         인플레이션 둔화에 대한 정부의 전망은?   \n",
       "7          미국 인플레이션 상황의 현재 상태는?   \n",
       "8       중동 리스크가 물가 상승에 미치는 영향은?   \n",
       "9             한국 금리 정책의 현재 전망은?   \n",
       "\n",
       "                                        simple_Q_ans  \\\n",
       "0         과일 가격 상승이 전체 인플레이션의 약 7분의 1을 차지하여 중요한 요소임.   \n",
       "1                      명절 이후에도 농산물 가격이 계속해서 상승하고 있음.   \n",
       "2           국제유가와 환율 상승으로 인해 물가상승률이 3%대에 이를 것으로 전망됨.   \n",
       "3        미국에서 서비스 물가 상승 지속으로 금리 인하 시기가 3월에서 6월로 미뤄짐.   \n",
       "4                           감귤, 사과, 배 등 과일 가격 크게 상승.   \n",
       "5                       국내 주유소의 휘발유·경유 가격이 3주 연속 상승.   \n",
       "6          정부는 소비자물가 상승률이 이달과 다음 달에 3%대에 진입할 것으로 전망.   \n",
       "7  미국은 인플레이션 둔화의 마지막 단계에서 어려움을 겪으며, 소비자물가지수와 도매물가...   \n",
       "8                  중동 리스크 장기화가 인플레이션 상승 위험을 높일 수 있음.   \n",
       "9  한국은행이 금리 인하 기대를 차단할 가능성이 커지며, 금리 인하 시점이 하반기로 밀...   \n",
       "\n",
       "                                            ldcc_ans  \\\n",
       "0  과일 가격 이 인플레이 션에 미치는  영향은 과일이 소비자들의  식료품 지출에 큰 ...   \n",
       "1  명절 이 지나 도 농산물  가격 은 여전히  고공행진 중이다 . 통계청에  따르면 ...   \n",
       "2  국제유가 와 환율  상승 은 수입 물 가격 이 오르 게 돼  소비 자물가에  반영되...   \n",
       "3  미국  중앙은행 (Fed)은 지난 달 금리 를 0.25%포인트   올린 데 이어  ...   \n",
       "4  한국 내 주요 물가 변화 에 대한  개요는 다음과  같습니다 .\\n\\n1. 국제 유...   \n",
       "5  최근  휘 발유 가격 은 하락하던  추세에서  다시  상승 세로  돌아섰다 . 한국...   \n",
       "6  기사 에 따르면  국제유가 와 원·달러  환율 의 동반 상승 에 “이달 물가상승률 ...   \n",
       "7  미국  인플레이션은  2022년 6월 이후  둔화되고  있는  추세이다 . 미국  ...   \n",
       "8  기사 에 따르면 서 중동 리스크가  물가 상승 에 미치는  영향에 대한  정확한 답...   \n",
       "9  기사 에 따르면  \"미국  ‘끈적한  서비스  물가’에 금리  인하 3월 → 6월로...   \n",
       "\n",
       "                                           llama_ans  \n",
       "0  귤값 2배 뛰고 휘발유값 꿈틀 ... 발목 잡힌 인플레 둔화, 고금리 길어지나.tx...  \n",
       "1  귤값 2배 뛰고 휘발유값 꿈틀 ... 발목 잡힌 인플레 둔화, 고금리 길어지나.tx...  \n",
       "2  1. 국제유가와 환율 상승은 수입 물가에 영향을 미쳐 국내 소비자 물가에 반영됩니다...  \n",
       "3  1. 국제유가와 원·달러 환율의 동반 상승으로 인플레이션이 다시 상승할 수 있다는 ...  \n",
       "4  귤값 2배 뛰고 휘발유값 꿈틀 ... 발목 잡힌 인플레 둔화, 고금리 길어지나.tx...  \n",
       "5  17일 한국석유공사 유가정보시스템 오피넷에 따르면 2월 둘째 주(11∼15일) 전국...  \n",
       "6  1. 인플레이션 둔화에 대한 정부의 전망은 현재 인플레이션이 둔화되고 있지만, 여전...  \n",
       "7  美 인플레이션 상황의 현재 상태는 둔화 속도가 느려지고 있으며, 국제유가와 원·달러...  \n",
       "8  1단계: 중동 리스크의 정의 이해\\n중동 리스크는 중동 지역의 지정학적 불안정, 정...  \n",
       "9  \\n한국의 금리 정책 전망은 현재 인플레이션이 둔화되고 있지만 여전히 높은 수준을 ...  "
      ]
     },
     "execution_count": 34,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df = pd.DataFrame({\n",
    "    \"simple_Q\": simple_questions,\n",
    "    \"simple_Q_ans\": simple_answers,\n",
    "    \"ldcc_ans\": ldcc_ans,\n",
    "    \"llama_ans\": llama_ans,\n",
    "})\n",
    "\n",
    "df.to_csv(\"simple_Q_local_models_comparison.csv\", index=False)\n",
    "\n",
    "df"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "cfda3e90-09a9-4868-b753-17df6f78a356",
   "metadata": {},
   "source": [
    "## ldcc,llama / advanced Q 답변 추론 /local"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "id": "c79208ee-fd98-4eec-9868-de66068cf19d",
   "metadata": {},
   "outputs": [],
   "source": [
    "questions_list = advanced_questions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "id": "74f6898b-3d54-42db-8f72-89623a9815f6",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "b564645950004dde90ccb1e3f3e80ead",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Loading checkpoint shards:   0%|          | 0/5 [00:00<?, ?it/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"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "답변 :  가계부채가 한국 의 경제 에 미치는  영향은 매우 심각합니다 . 가계부채가 증가할수 록 가계들은  소비 를 줄이고  저축 을 늘려야  합니다 . 이는  소비 가 국내 총생산 (GDP)의 약 60%를 차지하기  때문에  경제 의 성장 을 저해합니다 . 또한  가계들이  대출 을 갚 기 위해  이자 를 지불해야  하기  때문에 , 가계 소득 의 일부가  소비 에 사용되지  못하고  대출 을 갚 는 데 사용됩니다 . 이는  가계들의  생활 의 질을  떨 어뜨리고  경제 의 불평등 을 악 화시킬  수 있습니다 . 또한 가계부채가 너무  많 이 증가하면  은행 이 대출 을 꺼 려 대출 이 제한되고  이는  경제 의 성장 을 더욱  저\n",
      "답변 :  코로나 19 이후  소상공인의  경제 상황 은 악 화됐다 . 경기침 과 소비심리 의 위축 이 이어지면서 소상공인의 소득 은 이전으로  회복하지 못하고 있고 은행 대출금리 는 계속 오르 고 있다 . 그 결과  '디폴트(채무불이행)' 소상공인이 급증하고 있다 . 한국은행 의 집계에 따르면 2023년 9월 기준 인천 의 가계부채 총액은  108조1499억원이다 . 이중 주택담보대출 은 45조2306억원이다 . 문제 는 인천 이 인구대비  전세사기  피해 가 가장 많 이 발생 한 지역이고 , 역전세가  가장 심한  지역이라는  점이 다. 주택담보대출  만기 연장 시 가계부채는 더욱  늘어날  전망이 며, 가계부채가 \n",
      "답변 :  부자증세는  정부 에 추가 적인 세입을 확보할 수 있어  재정 을 강화할 수 있는  측면이  있다 . 그러나  부자증세는  부정적 인식 이 있을 수 있고  부자들의  투자 와 소비 를 저해할 수 있어  경제 에 부정적 영향을  미칠  수 있다 . 따라 서 부자증세의  실제 효과 와 부정적 영향은 세제 설계 와 구현 방식 에 따라  달라 질 수 있다 .\n",
      "\n",
      "소상공 인 지원 에 대한  영향은 부자증세의  세입이 어떻 게 사용되는 지에 따라  달라 질 수 있다 . 부자증세가  소상공 인 지원 을 위한  재정 을 확충하는  데 사용된다면  소상공 인 지원 에 긍 정적인  영향을  미칠  수 있다 . 그러나  \n",
      "답변 :  대출금리 가 오르면 대출금을  갚아야  하는  비용 이 늘어나 기 때문에 . 대출금을  갚 는 데 더 많은  돈 이 들어 가면 영업 에 쓸  돈 이 줄어들 고 결국 소상공인의  소득 이 줄어들 게 된다 . 이렇게  되면  은행  빚 을 갚 지 못하는  소상공인이  늘어나 게 된다 .\n",
      "\n",
      "대출금리 가 오르면 대출금을  갚아야  하는  비용 이 늘어나 기 때문에 . 대출금을  갚 는 데 더 많은  돈 이 들어 가면 영업 에 쓸  돈 이 줄어들 고 결국 소상공인의  소득 이 줄어들 게 된다 . 이렇게  되면  은행  빚 을 갚 지 못하는  소상공인이  늘어나 게 된다 . 대출금리 가 오르면 대출금을  갚아야  하는  비용 이 늘어나 기 때문에 .\n",
      "답변 :  가계부채 증가는  개인 과 지역  경제 에 많은  부정적인  영향을  끼친다 . 먼 저, 가계부채가 증가하면  가계들은  이자 와 원금을  갚느라  소비 를 줄이게  된다 . 이는  소비 가 경제 의 주요 성장  동력이라는  점에서  경제 의 성장 을 저해한다 .\n",
      "\n",
      "둘 째, 가계부채가 증가하면  가계들은  소득 의 많은  부분 을 빚 을 갚는데  사용하게  되어  저축 이 줄어들 고 경제 적 불안정성이  커진다 . 이는  가계들이  예기치 않은 상황 에 대처할 능력을  약화시키고  경제 적 불안정성을  증가시킨다 .\n",
      "\n",
      "셋 째, 가계부채가 증가하면  은행 은 대출 을 줄이기  위해  대출 의 기준 을 강화한다 . 이는  대출 이 필요한\n",
      "5\n"
     ]
    }
   ],
   "source": [
    "testing_models = \"LDCC/LDCC-SOLAR-10.7B\"\n",
    "embed_model = \"local:sentence-transformers/distiluse-base-multilingual-cased-v1\"\n",
    "\n",
    "ldcc_ans = []\n",
    "\n",
    "llm = HuggingFaceLLM(\n",
    "    model_name = testing_models,\n",
    "    tokenizer_name= testing_models,\n",
    "    query_wrapper_prompt=PromptTemplate(\"<s> [INST] {query_str} [/INST] \"),\n",
    "    model_kwargs={\"quantization_config\": bnb_config},\n",
    "    generate_kwargs={\"temperature\": 0.1,},\n",
    "    device_map=\"auto\",\n",
    "    )\n",
    "service_context = ServiceContext.from_defaults(embed_model=embed_model,llm = llm,chunk_size=1024)\n",
    "set_global_service_context(service_context)\n",
    "documents = SimpleDirectoryReader(\"./news2/\").load_data()\n",
    "index = VectorStoreIndex.from_documents(documents)\n",
    "engine = index.as_query_engine(response_mode=\"compact\",max_tokens=200)\n",
    "for question in questions_list:\n",
    "    message_input = (\n",
    "                f\"Thinking through several stages, answer politely in Korean. {question}\"\n",
    "            )\n",
    "    res = engine.query(message_input)\n",
    "    message = res.response\n",
    "    print(\"답변 : \", message)\n",
    "    ldcc_ans.append(message)\n",
    "\n",
    "print(len(ldcc_ans))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "id": "5b3ce389-f330-496c-b778-9f4d2de8c061",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "396f26df491342628df4771ae155ebd6",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Loading checkpoint shards:   0%|          | 0/3 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "답변 :  1단계: 가계부채의 정의와 규모 이해\n",
      "가계부채는 개인이 은행이나 기타 금융기관에서 빌린 돈을 의미합니다. 가계부채는 한국 경제에서 중요한 요소 중 하나로, 가계부채가 증가하면 소비와 투자에 영향을 미치고 경제 성장에 영향을 미칩니다.\n",
      "\n",
      "2단계: 가계부채가 한국 경제에 미치는 영향 분석\n",
      "가계부채가 증가하면 소비와 투자가 증가하여 경제 성장에 기여합니다. 그러나 부채가 과도하게 증가하면 가계의 부채 상환 능력이 감소하여 소비와 투자가 감소하고 경제 성장이 둔화됩니다. 또한, 가계부채가 증가하면 금융 시스템의 안정성이 감소하고 금융 기관이 대출을 회수할 때 경제 전반에 영향을 미칠 수 있습니다.\n",
      "\n",
      "3단계: 가계부채가 한국 경제에 미치는 부정적인 영향\n",
      "가계부채가 과도하게 증가하면 소비와 투자가 감소하여 경제 성장이 둔화됩니다. 또한, 가계부채가 증가하면 금융 시스템의 안정성이 감소하고 금융 기관이 대출을 회수할 때 경제 전반에 영향을 미칠 수 있습니다.\n",
      "\n",
      "4단계: 가계부채가 한국 경제에 미치는 긍정적인 영향\n",
      "가계부채가 증가하면 소비와 투자가 증가하여 경제 성장에 기여합니다. 또한, 가계부채는 은행과 기타 금융\n",
      "답변 :  COVID-19 이후 소상공인의 경제 상황은 악화되었습니다. 매출 감소와 물가 상승으로 인한 비용 증가로 인해 많은 소상공인이 빚을 내서 빚을 갚아야 했습니다. 또한, 은행 대출금리가 상승하면서 빚을 갚지 못하는 소상공인이 증가하고 있습니다. 이러한 상황은 가계부채 증가로 이어져 한국 경제의 뇌관이 되고 있습니다.\n",
      "답변 :  100대 기업 잉여금만 1000조 부자증세로 가계부채 덜자.txt\n",
      "\n",
      "부자증세는 정부 재정과 소상공인 지원에 여러 가지 영향을 미칠 수 있습니다.\n",
      "\n",
      "1. 정부 재정: 부자증세는 대기업의 잉여금을 활용하여 정부 재정을 확충하는 데 도움이 될 수 있습니다. 대기업의 잉여금이 증가하면 정부는 더 많은 세금을 거둘 수 있고, 이를 통해 사회복지, 교육, 인프라 등 다양한 분야에 대한 투자를 늘릴 수 있습니다.\n",
      "\n",
      "2. 소상공인 지원: 부자증세는 대기업의 잉여금을 활용하여 소상공인에게 대출을 제공하거나 재정 지원을 하는 데 사용할 수 있습니다. 대기업의 잉여금이 증가하면 정부는 소상공인에게 더 많은 대출을 제공할 수 있고, 소상공인에게 더 많은 재정 지원을 할 수 있습니다.\n",
      "\n",
      "3. 경제 성장: 부자증세는 대기업의 잉여금을 활용하여 경제 성장을 촉진하는 데 도움이 될 수 있습니다. 대기업의 잉여금이 증가하면 기업은 더 많은 투자를 할 수 있고, 이는 일자리 창출과 경제 성장에 기여할 수 있습니다.\n",
      "\n",
      "4. 사회 정의: 부자증세는 대기업의 잉여금을 활용하여 사회 정의를 증진하는 데 도움이 될 수 있습니다. 대기업의 잉여금이 증가하면 정부는 사회 복지, 교육\n",
      "답변 :  뱅크 대출 금리 상승은 소상공인에게 다음과 같은 영향을 미칩니다:\n",
      "\n",
      "1. 이자 부담 증가: 은행 대출금리가 상승하면 소상공인의 이자 부담이 증가합니다. 이는 소상공인의 수익에 직접적인 영향을 미칩니다.\n",
      "\n",
      "2. 부채 상환 능력 감소: 이자 부담이 증가하면 소상공인의 부채 상환 능력이 감소합니다. 이는 채무 불이행(디폴트)으로 이어질 수 있습니다.\n",
      "\n",
      "3. 소비 심리 위축: 이자 부담 증가로 인해 소상공인의 소비 심리가 위축되면 매출과 수익이 감소합니다. 이는 소상공인의 부채 상환 능력에 추가적인 영향을 미칩니다.\n",
      "\n",
      "4. 경제 전반에 미치는 영향: 소상공인 부채 상환 능력 감소는 은행의 대출 여력 감소로 이어져 경제 전반에 영향을 미칩니다. 이는 경제 성장률 저하, 실업률 증가, 소비 심리 위축으로 이어질 수 있습니다.\n",
      "답변 :  1단계: 질문 이해하기\n",
      "질문은 가계부채 증가가 개인과 지역 경제에 미치는 장단기적 영향을 묻는 질문입니다.\n",
      "\n",
      "2단계: 장단기 영향 분석하기\n",
      "가계부채 증가의 장단기 영향은 다음과 같습니다:\n",
      "\n",
      "장점:\n",
      "1. 가계부채 증가로 인해 소비가 증가하여 지역 경제에 활력을 불어넣을 수 있습니다.\n",
      "2. 가계부채 증가로 인해 은행 대출이 증가하여 지역 내 기업이나 프로젝트에 자금을 지원할 수 있습니다.\n",
      "\n",
      "단점:\n",
      "1. 가계부채 증가로 인해 이자 비용이 증가하여 가계의 소비 여력이 감소할 수 있습니다.\n",
      "2. 가계부채 증가로 인해 가계의 부채 상환 능력이 감소하여 디폴트(채무불이행)가 증가할 수 있습니다.\n",
      "3. 가계부채 증가로 인해 지역 내 부동산 가격이 상승하여 지역 내 자산 격차가 확대될 수 있습니다.\n",
      "\n",
      "3단계: 답변하기\n",
      "가계부채 증가의 장단기 영향은 개인과 지역 경제에 따라 달라질 수 있습니다. 가계부채 증가로 인해 소비가 증가하고 은행 대출이 증가하여 지역 경제에 활력을 불어넣을 수 있지만, 이자 비용이 증가하고 부채 상환 능력이 감소하여 디폴트가 증가할\n",
      "5\n"
     ]
    }
   ],
   "source": [
    "testing_models = \"mncai/llama2-13b-dpo-v7\"\n",
    "embed_model = \"local:sentence-transformers/distiluse-base-multilingual-cased-v1\"\n",
    "\n",
    "llama_ans = []\n",
    "\n",
    "llm = HuggingFaceLLM(\n",
    "    model_name = testing_models,\n",
    "    tokenizer_name= testing_models,\n",
    "    query_wrapper_prompt=PromptTemplate(\"<s> [INST] {query_str} [/INST] \"),\n",
    "    model_kwargs={\"quantization_config\": bnb_config},\n",
    "    generate_kwargs={\"temperature\": 0.1,},\n",
    "    device_map=\"auto\",\n",
    "    )\n",
    "service_context = ServiceContext.from_defaults(embed_model=embed_model,llm = llm,chunk_size=1024)\n",
    "set_global_service_context(service_context)\n",
    "documents = SimpleDirectoryReader(\"./news2/\").load_data()\n",
    "index = VectorStoreIndex.from_documents(documents)\n",
    "engine = index.as_query_engine(response_mode=\"compact\",max_tokens=200)\n",
    "for question in questions_list:\n",
    "    message_input = (\n",
    "                f\"Thinking through several stages, answer politely in Korean. {question}\"\n",
    "            )\n",
    "    res = engine.query(message_input)\n",
    "    message = res.response\n",
    "    print(\"답변 : \", message)\n",
    "    llama_ans.append(message)\n",
    "\n",
    "print(len(llama_ans))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "id": "dd43207e-f13c-420f-93fe-378befd49a9b",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>advanced_Q</th>\n",
       "      <th>advanced_Q_ans</th>\n",
       "      <th>ldcc_ans</th>\n",
       "      <th>llama_ans</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>가계부채가 한국 경제에 미치는 영향에 대해 설명하세요.</td>\n",
       "      <td>가계부채 증가는 개인의 소비를 억제하고, 금리 상승 시 이자 부담 증가로 이어져 가...</td>\n",
       "      <td>가계부채가 한국 의 경제 에 미치는  영향은 매우 심각합니다 . 가계부채가 증가할수...</td>\n",
       "      <td>1단계: 가계부채의 정의와 규모 이해\\n가계부채는 개인이 은행이나 기타 금융기관에서...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>코로나19 이후 소상공인의 경제 상황 변화는 어떠한가요?</td>\n",
       "      <td>코로나19 이후 소상공인은 매출 감소와 빚 증가로 인해 경제적 어려움을 겪고 있으며...</td>\n",
       "      <td>코로나 19 이후  소상공인의  경제 상황 은 악 화됐다 . 경기침 과 소비심리 의...</td>\n",
       "      <td>COVID-19 이후 소상공인의 경제 상황은 악화되었습니다. 매출 감소와 물가 상승...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>부자증세가 정부 재정과 소상공인 지원에 어떻게 영향을 미칠까요?</td>\n",
       "      <td>부자증세를 통한 추가 재정 확보는 정부와 지자체가 소상공인 지원과 같은 사회 안전망...</td>\n",
       "      <td>부자증세는  정부 에 추가 적인 세입을 확보할 수 있어  재정 을 강화할 수 있는 ...</td>\n",
       "      <td>100대 기업 잉여금만 1000조 부자증세로 가계부채 덜자.txt\\n\\n부자증세는 ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>은행 대출금리 상승이 소상공인에게 미치는 영향은 무엇인가요?</td>\n",
       "      <td>은행 대출금리 상승은 소상공인의 대출 이자 부담을 증가시켜 재정적 어려움을 가중시키...</td>\n",
       "      <td>대출금리 가 오르면 대출금을  갚아야  하는  비용 이 늘어나 기 때문에 . 대출금...</td>\n",
       "      <td>뱅크 대출 금리 상승은 소상공인에게 다음과 같은 영향을 미칩니다:\\n\\n1. 이자 ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>가계부채 증가가 개인과 지역 경제에 끼치는 장단기적 영향은 무엇인가요?</td>\n",
       "      <td>가계부채 증가는 단기적으로는 생활비 마련을 위한 대출 의존도를 높이지만, 장기적으로...</td>\n",
       "      <td>가계부채 증가는  개인 과 지역  경제 에 많은  부정적인  영향을  끼친다 . 먼...</td>\n",
       "      <td>1단계: 질문 이해하기\\n질문은 가계부채 증가가 개인과 지역 경제에 미치는 장단기적...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                advanced_Q  \\\n",
       "0           가계부채가 한국 경제에 미치는 영향에 대해 설명하세요.   \n",
       "1          코로나19 이후 소상공인의 경제 상황 변화는 어떠한가요?   \n",
       "2      부자증세가 정부 재정과 소상공인 지원에 어떻게 영향을 미칠까요?   \n",
       "3        은행 대출금리 상승이 소상공인에게 미치는 영향은 무엇인가요?   \n",
       "4  가계부채 증가가 개인과 지역 경제에 끼치는 장단기적 영향은 무엇인가요?   \n",
       "\n",
       "                                      advanced_Q_ans  \\\n",
       "0  가계부채 증가는 개인의 소비를 억제하고, 금리 상승 시 이자 부담 증가로 이어져 가...   \n",
       "1  코로나19 이후 소상공인은 매출 감소와 빚 증가로 인해 경제적 어려움을 겪고 있으며...   \n",
       "2  부자증세를 통한 추가 재정 확보는 정부와 지자체가 소상공인 지원과 같은 사회 안전망...   \n",
       "3  은행 대출금리 상승은 소상공인의 대출 이자 부담을 증가시켜 재정적 어려움을 가중시키...   \n",
       "4  가계부채 증가는 단기적으로는 생활비 마련을 위한 대출 의존도를 높이지만, 장기적으로...   \n",
       "\n",
       "                                            ldcc_ans  \\\n",
       "0  가계부채가 한국 의 경제 에 미치는  영향은 매우 심각합니다 . 가계부채가 증가할수...   \n",
       "1  코로나 19 이후  소상공인의  경제 상황 은 악 화됐다 . 경기침 과 소비심리 의...   \n",
       "2  부자증세는  정부 에 추가 적인 세입을 확보할 수 있어  재정 을 강화할 수 있는 ...   \n",
       "3  대출금리 가 오르면 대출금을  갚아야  하는  비용 이 늘어나 기 때문에 . 대출금...   \n",
       "4  가계부채 증가는  개인 과 지역  경제 에 많은  부정적인  영향을  끼친다 . 먼...   \n",
       "\n",
       "                                           llama_ans  \n",
       "0  1단계: 가계부채의 정의와 규모 이해\\n가계부채는 개인이 은행이나 기타 금융기관에서...  \n",
       "1  COVID-19 이후 소상공인의 경제 상황은 악화되었습니다. 매출 감소와 물가 상승...  \n",
       "2  100대 기업 잉여금만 1000조 부자증세로 가계부채 덜자.txt\\n\\n부자증세는 ...  \n",
       "3  뱅크 대출 금리 상승은 소상공인에게 다음과 같은 영향을 미칩니다:\\n\\n1. 이자 ...  \n",
       "4  1단계: 질문 이해하기\\n질문은 가계부채 증가가 개인과 지역 경제에 미치는 장단기적...  "
      ]
     },
     "execution_count": 39,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df = pd.DataFrame({\n",
    "    \"advanced_Q\": advanced_questions,\n",
    "    \"advanced_Q_ans\": advanced_answers,\n",
    "    \"ldcc_ans\": ldcc_ans,\n",
    "    \"llama_ans\": llama_ans,\n",
    "})\n",
    "\n",
    "df.to_csv(\"advanced_Q_local_models_comparison.csv\", index=False)\n",
    "\n",
    "df"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4504356a-829b-475d-83b1-5572a3b9cb95",
   "metadata": {},
   "source": [
    "## cosine 유사도 계산"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "id": "01f04511-83fc-4fb0-b664-ac8b86071fbd",
   "metadata": {},
   "outputs": [],
   "source": [
    "from sentence_transformers import SentenceTransformer\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "from sklearn.metrics.pairwise import cosine_similarity as sklearn_cosine_similarity\n",
    "\n",
    "data_path = 'advanced_Q_gpt_models_comparison.csv'\n",
    "data = pd.read_csv(data_path)\n",
    "# Load the pre-trained SentenceTransformer model\n",
    "model = SentenceTransformer('distiluse-base-multilingual-cased-v1')\n",
    "\n",
    "# Function to calculate cosine similarity\n",
    "def calculate_similarity(text_list1, text_list2):\n",
    "    embeddings1 = model.encode(text_list1, convert_to_tensor=True)\n",
    "    embeddings2 = model.encode(text_list2, convert_to_tensor=True)\n",
    "    \n",
    "    # Move embeddings to CPU if they are on GPU\n",
    "    if embeddings1.is_cuda:\n",
    "        embeddings1 = embeddings1.cpu()\n",
    "    if embeddings2.is_cuda:\n",
    "        embeddings2 = embeddings2.cpu()\n",
    "    \n",
    "    # Convert PyTorch tensors to NumPy arrays\n",
    "    embeddings1_np = embeddings1.numpy()\n",
    "    embeddings2_np = embeddings2.numpy()\n",
    "    \n",
    "    # Use sklearn's cosine_similarity\n",
    "    similarity_scores = sklearn_cosine_similarity(embeddings1_np, embeddings2_np)\n",
    "    \n",
    "    # Return diagonal elements which represent the similarity scores between corresponding vectors\n",
    "    return np.diag(similarity_scores)\n",
    "\n",
    "# Calculate cosine similarity scores\n",
    "gpt4_similarity_scores = calculate_similarity(data['advanced_Q_ans'].tolist(), data['GPT-4 Turbo'].tolist())\n",
    "gpt35_similarity_scores = calculate_similarity(data['advanced_Q_ans'].tolist(), data['GPT-3.5 Turbo'].tolist())\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "id": "0408f986-18ef-4475-becc-10e38f776d81",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1200x600 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "question_numbers = np.arange(len(data))  # Question numbers for X-axis\n",
    "\n",
    "plt.figure(figsize=(12, 6))\n",
    "bar_width = 0.2  # Width of the bars\n",
    "\n",
    "# Set position of bar on X axis\n",
    "r1 = np.arange(len(simulated_scores_gpt4))\n",
    "r2 = [x + bar_width for x in r1]\n",
    "\n",
    "# Make the plot with pastel colors\n",
    "plt.bar(r1, simulated_scores_gpt4, color='skyblue', width=bar_width, edgecolor='grey', label='GPT-4 Turbo')\n",
    "plt.bar(r2, simulated_scores_gpt35, color='lightpink', width=bar_width, edgecolor='grey', label='GPT-3.5 Turbo')\n",
    "\n",
    "# Adjust xticks to be in the middle of the group bars\n",
    "plt.xlabel('Question', fontweight='bold')\n",
    "plt.xticks([r + bar_width / 2 for r in range(len(simulated_scores_gpt4))], question_numbers)\n",
    "plt.ylabel('Cosine Similarity wrt Answer')\n",
    "plt.title('advanced_Q Comparing gpt models, TASK : RAG')\n",
    "plt.grid(axis='y') \n",
    "# Create legend & Show graphic\n",
    "plt.legend()\n",
    "plt.show()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "id": "721ad4d5-fd76-46f4-9cf6-c7e6950f9801",
   "metadata": {},
   "outputs": [],
   "source": [
    "data_path = 'advanced_Q_local_models_comparison.csv'\n",
    "data = pd.read_csv(data_path)\n",
    "\n",
    "ldcc_similarity_scores = calculate_similarity(data['advanced_Q_ans'].tolist(), data['ldcc_ans'].tolist())\n",
    "llama_similarity_scores = calculate_similarity(data['advanced_Q_ans'].tolist(), data['llama_ans'].tolist())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "id": "42ebd657-6d7f-4094-af57-de360ddc6053",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1200x600 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "question_numbers = np.arange(len(data))  # Question numbers for X-axis\n",
    "\n",
    "plt.figure(figsize=(12, 6))\n",
    "bar_width = 0.2  # Width of the bars\n",
    "\n",
    "# Set position of bar on X axis\n",
    "r1 = np.arange(len(simulated_scores_gpt4))\n",
    "r2 = [x + bar_width for x in r1]\n",
    "\n",
    "# Make the plot with pastel colors\n",
    "plt.bar(r1, ldcc_similarity_scores, color='skyblue', width=bar_width, edgecolor='grey', label='ldcc')\n",
    "plt.bar(r2, llama_similarity_scores, color='lightpink', width=bar_width, edgecolor='grey', label='llama')\n",
    "\n",
    "# Adjust xticks to be in the middle of the group bars\n",
    "plt.xlabel('Question', fontweight='bold')\n",
    "plt.xticks([r + bar_width / 2 for r in range(len(simulated_scores_gpt4))], question_numbers)\n",
    "plt.ylabel('Cosine Similarity wrt Answer')\n",
    "plt.title('advanced_Q Comparing local models, TASK : RAG')\n",
    "plt.grid(axis='y') \n",
    "# Create legend & Show graphic\n",
    "plt.legend()\n",
    "plt.show()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "id": "e825acf7-10aa-4394-91c4-94a67f26e987",
   "metadata": {},
   "outputs": [],
   "source": [
    "data_path = 'simple_Q_gpt_models_comparison.csv'\n",
    "data = pd.read_csv(data_path)\n",
    "\n",
    "simulated_scores_gpt4 = calculate_similarity(data['simple_Q_ans'].tolist(), data['GPT-4 Turbo'].tolist())\n",
    "simulated_scores_gpt35 = calculate_similarity(data['simple_Q_ans'].tolist(), data['GPT-3.5 Turbo'].tolist())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "id": "ab7fe05d-11e1-4630-ab96-65c849c01722",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1200x600 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "question_numbers = np.arange(len(data))  # Question numbers for X-axis\n",
    "\n",
    "plt.figure(figsize=(12, 6))\n",
    "bar_width = 0.2  # Width of the bars\n",
    "\n",
    "# Set position of bar on X axis\n",
    "r1 = np.arange(len(simulated_scores_gpt4))\n",
    "r2 = [x + bar_width for x in r1]\n",
    "\n",
    "# Make the plot with pastel colors\n",
    "plt.bar(r1, simulated_scores_gpt4, color='skyblue', width=bar_width, edgecolor='grey', label='GPT-4 Turbo')\n",
    "plt.bar(r2, simulated_scores_gpt35, color='lightpink', width=bar_width, edgecolor='grey', label='GPT-3.5 Turbo')\n",
    "\n",
    "# Adjust xticks to be in the middle of the group bars\n",
    "plt.xlabel('Question', fontweight='bold')\n",
    "plt.xticks([r + bar_width / 2 for r in range(len(simulated_scores_gpt4))], question_numbers)\n",
    "plt.ylabel('Cosine Similarity wrt Answer')\n",
    "plt.title('simple_Q Comparing gpt models, TASK : RAG')\n",
    "plt.grid(axis='y') \n",
    "# Create legend & Show graphic\n",
    "plt.legend()\n",
    "plt.show()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "aaf247e1-65ad-4708-b94b-d8a698d0383e",
   "metadata": {},
   "outputs": [],
   "source": [
    "average_score_gpt4 = np.mean(simulated_scores_gpt4)\n",
    "average_score_gpt35 = np.mean(simulated_scores_gpt35)\n",
    "\n",
    "# Extend the question numbers array to include a slot for the average\n",
    "extended_question_numbers = np.append(question_numbers, [len(question_numbers)])  # Add an index for the average\n",
    "\n",
    "plt.figure(figsize=(14, 7))\n",
    "bar_width = 0.2  # Width of the bars\n",
    "\n",
    "# Set position of bar on X axis for the extended array\n",
    "r1_extended = np.arange(len(extended_question_numbers))\n",
    "r2_extended = [x + bar_width for x in r1_extended]\n",
    "\n",
    "# Plot the original scores\n",
    "plt.bar(r1, simulated_scores_gpt4, color='skyblue', width=bar_width, edgecolor='grey', label='GPT-4 Turbo')\n",
    "plt.bar(r2, simulated_scores_gpt35, color='lightpink', width=bar_width, edgecolor='grey', label='GPT-3.5 Turbo')\n",
    "\n",
    "# Plot the average scores as the last pair of bars\n",
    "plt.bar(len(extended_question_numbers)-1 - 0.2, average_score_gpt4, color='skyblue', width=bar_width, edgecolor='grey')\n",
    "plt.bar(len(extended_question_numbers)-1, average_score_gpt35, color='lightpink', width=bar_width, edgecolor='grey')\n",
    "\n",
    "# Adjust xticks to include the average\n",
    "plt.xlabel('Question', fontweight='bold')\n",
    "plt.xticks(r1_extended + bar_width / 2, list(question_numbers) + ['Average'])\n",
    "plt.ylabel('Simulated Cosine Similarity')\n",
    "plt.title('simeple_Q Comparing local models, TASK : RAG (Including Average)')\n",
    "plt.grid(axis='y')\n",
    "\n",
    "# Create legend & Show graphic\n",
    "plt.legend()\n",
    "plt.show()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "id": "fff97f63-d56f-49ee-8d1b-fcce7295c052",
   "metadata": {},
   "outputs": [],
   "source": [
    "data_path = 'simple_Q_local_models_comparison.csv'\n",
    "data = pd.read_csv(data_path)\n",
    "\n",
    "ldcc_similarity_scores = calculate_similarity(data['simple_Q_ans'].tolist(), data['ldcc_ans'].tolist())\n",
    "llama_similarity_scores = calculate_similarity(data['simple_Q_ans'].tolist(), data['llama_ans'].tolist())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "id": "eb0599e9-cb63-4c38-a9e7-786fe58cb476",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAA+kAAAIjCAYAAAB/OVoZAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjguMiwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy8g+/7EAAAACXBIWXMAAA9hAAAPYQGoP6dpAABulUlEQVR4nO3deZyN9f//8eeZfTELmcU6k33JLj4IU2REig8lxBilhal8JooWS8lS0RQTpQ9KRFnqk6QylhZkCylLskbGCDPMYbZz/f7wm/PtNIM5M2fmHDOP++12brc57+u63tfrep/T5DnX8jYZhmEIAAAAAAA4nZuzCwAAAAAAAFcQ0gEAAAAAcBGEdAAAAAAAXAQhHQAAAAAAF0FIBwAAAADARRDSAQAAAABwEYR0AAAAAABcBCEdAAAAAAAXQUgHAAAAAMBFENIBoIgGDx6syMhIZ5dxXSaTSePHj3d2GbiOI0eOyGQyaf78+SW+7/Xr18tkMmn9+vUlvm9XrON6ivJZ3SjHCAAoeYR0AIBDGIahBQsWqEOHDgoODpafn58aNWqkiRMnymw229XXzp079eCDD6patWry9vZWhQoV1LlzZ82bN085OTnFdARA6RAZGSmTyXTd19//uLBq1SqZTCZVrlxZFosl334zMzP15ptvqlmzZgoMDFRwcLAaNmyoRx55RPv27bOuN3/+fJlMJm3bts1m+9TUVLVq1Uo+Pj5avXq1Q441948duS93d3eFhoaqT58+2rt371W3K8jxSlJaWppeeeUVtWzZUkFBQfL29lZERIT69u2rL774wiHHAAD/5OHsAgDgRjdnzpxr/iOvLMjJyVH//v318ccfq3379ho/frz8/Pz03Xffady4cfr444+1Zs0ahYaGXrev9957T4899pjCwsI0cOBA1a5dWxcuXFBSUpIeeugh/fnnn3ruuedK4KicIyIiQpcuXZKnp6ezS8ENKiEhQRcvXrS+X7VqlT766CO98cYbqlixorW9bdu21p8XLlyoyMhIHTlyRGvXrlXnzp3z9Nu7d299+eWX6tevn4YOHaqsrCzt27dPK1euVNu2bVWvXr2r1pSWlqYuXbpo9+7dWrFihbp27eqgo73iySef1K233qqsrCzt3r1bs2fP1vr167Vnzx6Fh4fnWb8gx3vw4EFFR0fr6NGj6tWrlwYNGqRy5crp+PHjWrVqle6++2598MEHGjhwoEOPBQAI6QBQRIQp6dVXX9XHH3+skSNH6rXXXrO2P/LII7r//vvVs2dPxcbGXvfM0+bNm/XYY4+pTZs2WrVqlQICAqzLRowYoW3btmnPnj3FdhzOlJ2dLYvFIi8vL/n4+Di7HNzAevbsafP+1KlT+uijj9SzZ898b81JT0/XZ599psmTJ2vevHlauHBhntC6detWrVy5Uq+88kqeP5LNnDlT58+fv2o9Fy5cUHR0tHbu3Knly5frrrvuKuyhXVX79u3Vp08f6/u6devq8ccf1wcffKBnnnnGZt2CHG92drZ69eql5ORkbdiwQe3atbNZPm7cOH399ddc2QOgWHC5OwBcw4ULFzRixAhFRkbK29tboaGhuvPOO7Vjxw7rOv+8Jz33PtXXX39diYmJqlGjhvz8/NSlSxcdP35chmHo5ZdfVtWqVeXr66t7771XZ8+ezbPvL7/8Uu3bt5e/v78CAgLUvXt3/fLLLzbrDB48WOXKldOhQ4cUHR0tf39/Va5cWS+99JIMw7ju8Z04cUJDhgxRWFiYvL291bBhQ82dO9euMbp06ZJee+011alTR5MnT86zvEePHoqJidGqVau0ZcuWa/Y1YcIEmUwmLVy40Cag52rZsqUGDx5sfZ+enq6nn37aell83bp19frrr+c5dpPJpLi4OH3yySdq0KCBfH191aZNG/3888+SpHfeeUe1atWSj4+PoqKidOTIEZvto6KidMstt2j79u1q27atfH19dfPNN2v27Nk262VmZmrs2LFq0aKFgoKC5O/vr/bt22vdunU26/39O5KQkKCaNWvK29tbv/76a773Oed+zidOnFDPnj1Vrlw5hYSEaOTIkXlCwl9//aWBAwdaL0eOiYnRrl27inSf+yeffKIWLVrI19dXFStW1IMPPqgTJ07kWW/fvn26//77FRISIl9fX9WtW1fPP/+8dfnRo0c1bNgw1a1bV76+vrrpppt033335Rnvgho/frxMJpMOHDigBx98UEFBQQoJCdGLL74owzB0/Phx3XvvvQoMDFR4eLimTZuWp4/Tp0/roYceUlhYmHx8fNSkSRO9//77edY7f/68Bg8erKCgIOu4Xi2Y7tu3T3369FGFChXk4+Ojli1b6n//+991j+e3335T7969FR4eLh8fH1WtWlUPPPCAUlNT7R4be6xYsUKXLl3SfffdpwceeEDLly/X5cuXbdb5/fffJSlPWJUkd3d33XTTTfn2ffHiRXXt2lU7duzQsmXL1L179wLV9Pvvv1v3WRjt27e39vNPBTneTz75RHv27NGLL76Y7zFLUpcuXYrlDw4AQEgHgGt47LHHNGvWLPXu3Vtvv/22Ro4cKV9f32ve65hr4cKFevvtt/XEE0/o6aef1oYNG3T//ffrhRde0OrVq/Xss8/qkUce0eeff66RI0fabLtgwQJ1795d5cqV09SpU/Xiiy/q119/1W233ZYn0OTk5Khr164KCwvTq6++qhYtWmjcuHEaN27cNetLTk7Wv/71L61Zs0ZxcXF68803VatWLT300ENKSEgo8Bh9//33OnfunPr37y8Pj/wv0Bo0aJAk6fPPP79qP2azWUlJSerQoYOqV69+3f0ahqF77rlHb7zxhrp27arp06erbt26GjVqlOLj4/Os/9133+npp59WTEyMxo8fr7179+ruu+9WYmKi3nrrLQ0bNkyjRo3Spk2bNGTIkDzbnzt3Tt26dVOLFi306quvqmrVqnr88cdt/qiRlpam9957T1FRUZo6darGjx+vlJQU61nEf5o3b55mzJihRx55RNOmTVOFChWuerw5OTmKjo7WTTfdpNdff10dO3bUtGnT9O6771rXsVgs6tGjhz766CPFxMTolVde0Z9//qmYmJjrjufVzJ8/X/fff7/c3d01efJkDR06VMuXL9dtt91mE1J3796t1q1ba+3atRo6dKjefPNN9ezZ0+Yz37p1qzZu3KgHHnhAb731lh577DElJSUpKirK7ucW/F3fvn1lsVg0ZcoUtW7dWhMnTlRCQoLuvPNOValSRVOnTlWtWrU0cuRIffvtt9btLl26pKioKC1YsEADBgzQa6+9pqCgIA0ePFhvvvmmdT3DMHTvvfdqwYIFevDBBzVx4kT98ccf+Y7rL7/8on/961/au3evRo8erWnTpsnf3189e/bUihUrrnoMmZmZio6O1ubNm/XEE08oMTFRjzzyiA4dOnTNs9SOsHDhQt1+++0KDw/XAw88oAsXLuT5bzUiIsK6bnZ2doH6TU9P11133aWtW7fqk08+0d13313gmjp16qROnToV/CD+Iff3ZPny5fMsK8jx5r5/8MEHC10DABSaAQC4qqCgIGP48OHXXCcmJsaIiIiwvj98+LAhyQgJCTHOnz9vbR8zZowhyWjSpImRlZVlbe/Xr5/h5eVlXL582TAMw7hw4YIRHBxsDB061GY/p06dMoKCgmzaY2JiDEnGE088YW2zWCxG9+7dDS8vLyMlJcXaLskYN26c9f1DDz1kVKpUyThz5ozNfh544AEjKCjIMJvN1zzuXAkJCYYkY8WKFVdd5+zZs4Yk49///vdV19m1a5chyXjqqacKtN9PP/3UkGRMnDjRpr1Pnz6GyWQyDh48aG2TZHh7exuHDx+2tr3zzjuGJCM8PNxIS0uztud+Tn9ft2PHjoYkY9q0ada2jIwMo2nTpkZoaKiRmZlpGIZhZGdnGxkZGTb1nDt3zggLCzOGDBlibcv9jgQGBhqnT5+2WT932bx586xtuZ/zSy+9ZLNus2bNjBYtWljfL1u2zJBkJCQkWNtycnKMO+64I0+f+Vm3bp0hyVi3bp1hGIaRmZlphIaGGrfccotx6dIl63orV640JBljx461tnXo0MEICAgwjh49atOnxWKx/pzfd2rTpk2GJOODDz64ah1XM27cOEOS8cgjj1jbsrOzjapVqxomk8mYMmWKtf3cuXOGr6+vERMTY23L/e5++OGH1rbMzEyjTZs2Rrly5azfi9zv2quvvmqzn/bt2+cZ106dOhmNGjWy/vecOwZt27Y1ateufdVj/OmnnwxJxieffHLNYy6M1157Lc93OldycrLh4eFhzJkzx9rWtm1b495777VZz2KxWP87CAsLM/r162ckJibm+bwNwzDmzZtnSDIiIiIMT09P49NPP7W75oiICJvfq1eTO45z5841UlJSjJMnTxqrV682atWqZZhMJmPLli2FOt5mzZoZwcHBefZ38eJFIyUlxfpKTU21+9gA4Ho4kw4A1xAcHKwff/xRJ0+etHvb++67T0FBQdb3rVu3lnTlzMzfzzi3bt1amZmZ1suHv/nmG50/f179+vXTmTNnrC93d3e1bt06z6XTkhQXF2f9OffS7szMTK1Zsybf2gzD0LJly9SjRw8ZhmGzn+joaKWmptpc0n8tFy5ckKR8L0/Plbssd938pKWlXbefv1u1apXc3d315JNP2rQ//fTTMgxDX375pU17p06dbG5LyP08evfubbPP3PZDhw7ZbO/h4aFHH33U+t7Ly0uPPvqoTp8+re3bt0u6ctmvl5eXpCtntc+ePavs7Gy1bNky3/Hs3bu3QkJCCnS80pUrO/6uffv2NnWuXr1anp6eGjp0qLXNzc1Nw4cPL/A+/m7btm06ffq0hg0bZnOffPfu3VWvXj3rMwZSUlL07bffasiQIXmugjCZTNaffX19rT9nZWXpr7/+Uq1atRQcHFzg71t+Hn74YevP7u7uatmypQzD0EMPPWRtDw4OVt26dW3Ga9WqVQoPD1e/fv2sbZ6ennryySd18eJFbdiwwbqeh4eHHn/8cZv9PPHEEzZ1nD17VmvXrtX999+vCxcuWP+b+uuvvxQdHa3ffvst39sEJFl/V3z11VdFuqrAXosXL5abm5t69+5tbevXr5++/PJLnTt3ztpmMpn01VdfaeLEiSpfvrw++ugjDR8+3Pqk8/zO9icnJ8vHx0fVqlWzu64jR47YdRvEkCFDFBISosqVK6tr165KTU3VggULdOuttxbqeNPS0lSuXLk8+3n++ecVEhJiffXv39/uYwOA6yGkA8A1vPrqq9qzZ4+qVaumVq1aafz48XnC29X8M6zk/iP8n/9gzW3P/Qfib7/9Jkm64447bP4xGBISoq+//lqnT5+22d7NzU01atSwaatTp44kXfUfuSkpKTp//rzefffdPPuIjY2VpDz7uZqCBPDcZdd6untgYOB1+/m7o0ePqnLlynlCff369a3L/66wn0euypUry9/f36Ytv3F+//331bhxY/n4+Oimm25SSEiIvvjii3zvK7755puveYx/5+PjkyfQly9f3qbOo0ePqlKlSvLz87NZr1atWgXez9/ljmHdunXzLKtXr551ee5/E7fccss1+7t06ZLGjh1rfYZAxYoVFRISovPnzxfpvuv8PlsfHx+bJ5nntv9zvGrXri03N9t/Dv3zO5Q7rv8Mbf8cl4MHD8owDL344ot5/rvKvf3kav9d3XzzzYqPj9d7772nihUrKjo6WomJicV+P/qHH36oVq1a6a+//tLBgwd18OBBNWvWTJmZmfrkk09s1vX29tbzzz+vvXv36uTJk/roo4/0r3/9Sx9//LHNHwpzvfPOO/Ly8lLXrl21f//+Yj2OsWPH6ptvvtGKFSs0aNAgpaam5vlcpYIfb0BAgM0T8nMNGzZM33zzjb755huFhYUV6zEBKLt4ujsAXMP999+v9u3ba8WKFfr666/12muvaerUqQV6QrG7u7td7cb/f9hZ7nRuCxYsyHfqoKvd922P3H08+OCDV71fuXHjxgXqq0GDBpKu3JP8z6dK59q9e7ck5fljwt/VqlVLHh4e1oe5OVphPw97fPjhhxo8eLB69uypUaNGKTQ01Hovd34PsPr7meXruVqdN5InnnhC8+bN04gRI9SmTRsFBQXJZDLpgQceKNI0hvmNjSM/14LKPYaRI0cqOjo633Wu9QeTadOmafDgwfrss8/09ddf68knn9TkyZO1efNmVa1a1eH1/vbbb9q6daskqXbt2nmWL1y4UI888ki+21aqVEkPPPCAevfurYYNG+rjjz/W/PnzbX4/NWjQQKtWrVKnTp1055136ocffijUWfWCaNSokfUJ7T179pTZbNbQoUN12223Wfdpz/HWq1dPO3fu1IkTJ1SlShXrOnXq1LH+cY5ZGAAUF0I6AFxHpUqVNGzYMA0bNkynT59W8+bN9corrxTbU31r1qwp6cpZ5/zm7v0ni8WiQ4cOWf/hKEkHDhyQpHynW5KkkJAQBQQEKCcnp0D7uJZ27dopODhYixYt0vPPP59vOPrggw8kXbkF4Gr8/Px0xx13aO3atTp+/Ph1/zEfERGhNWvW6MKFCzZn0/ft22dd7kgnT55Uenq6zdn0f47z0qVLVaNGDS1fvtzmMu/rPcTPUSIiIrRu3TqZzWabs+kHDx4sdH+StH//ft1xxx02y/bv329dnvvHl+tNj7d06VLFxMTYPGX98uXLxf5gtKuJiIjQ7t27ZbFYbM66/vM7FBERoaSkJF28eNHmbPo/zw7njoOnp2eh/7tq1KiRGjVqpBdeeEEbN25Uu3btNHv2bE2cOLFQ/V3LwoUL5enpqQULFuT57/b777/XW2+9pWPHjl3zQY6enp5q3LixfvvtN505cybPHxZbtWqlTz/9VN27d9edd96p7777zq5bPAprypQpWrFihV555RXrLAz2HO/dd9+txYsXa+HChXmmcAOA4sbl7gBwFTk5OXkuNQ0NDVXlypWVkZFRbPuNjo5WYGCgJk2apKysrDzLU1JS8rTNnDnT+rNhGJo5c6Y8PT2v+nRkd3d39e7dW8uWLcs3WOW3j6vx8/PTM888o/3799tMt5Xriy++0Pz589WjRw81atTomn2NGzdOhmFo4MCB+V5qun37duv0WN26dVNOTo7NsUvSG2+8IZPJ5PA/omRnZ+udd96xvs/MzNQ777yjkJAQtWjRQtL/nb39+9naH3/8UZs2bXJoLVcTHR2trKwszZkzx9pmsViUmJhYqP5atmyp0NBQzZ492+Y7/+WXX2rv3r3W6bRCQkLUoUMHzZ07V8eOHbPp4+9j4e7unudM9owZM5w213S3bt106tQpLVmyxNqWnZ2tGTNmqFy5curYsaN1vezsbM2aNcu6Xk5OjmbMmGHTX2hoqKKiovTOO+/ozz//zLO/a/13lZaWluep6Y0aNZKbm1ux/b5ZuHCh2rdvr759+6pPnz42r1GjRkmSPvroI0lXzkL/87OVrkxNt2nTJpUvX/6q4btTp0766KOPdPDgQXXt2tX6/IlrKeoUbDVr1lTv3r01f/58nTp1SpJ9x3v//ferQYMGevnll7V58+Z891GcV2UAKNs4kw4AV3HhwgVVrVpVffr0UZMmTVSuXDmtWbNGW7duzXe+ZUcJDAzUrFmzNHDgQDVv3lwPPPCAQkJCdOzYMX3xxRdq166dTTD18fHR6tWrFRMTo9atW+vLL7/UF198oeeee+6aZ6ymTJmidevWqXXr1ho6dKgaNGigs2fPaseOHVqzZk2+c7dfzTPPPKOdO3dq6tSp2rRpk3r37i1fX199//33+vDDD9WwYcMCzdHdtm1bJSYmatiwYapXr54GDhyo2rVr68KFC1q/fr3+97//Wc8o9ujRQ7fffruef/55HTlyRE2aNNHXX3+tzz77TCNGjLBekeAolStX1tSpU3XkyBHVqVNHS5Ys0c6dO/Xuu+/K09NTknT33Xdr+fLl6tWrl7p3767Dhw9r9uzZatCgQb5/dHC0nj17qlWrVnr66ad18OBB1atXT//73/+sn+Xfz+4XhKenp6ZOnarY2Fh17NhR/fr1U3Jyst58801FRkbqP//5j3Xdt956S7fddpuaN2+uRx55RDfffLOOHDmiL774wjr93N13360FCxYoKChIDRo00KZNm7RmzZqrzrFd3B555BG98847Gjx4sLZv367IyEgtXbpUP/zwgxISEqxXaPTo0UPt2rXT6NGjdeTIETVo0EDLly/P937xxMRE3XbbbWrUqJGGDh2qGjVqKDk5WZs2bdIff/yhXbt25VvL2rVrFRcXp/vuu0916tRRdna29Yzv3x9yNn78eE2YMEHr1q1TVFRUoY/9xx9/1MGDB/O9l1ySqlSpoubNm2vhwoV69tlntWvXLvXv31933XWX2rdvrwoVKujEiRN6//33dfLkSSUkJFzzloxevXppzpw5GjJkiO655x6tXr36mpeL5/6B0Z6Hx/3TqFGj9PHHHyshIUG9evWy63g9PT21YsUKRUdH67bbbtO///1vtW/fXv7+/jpx4oT+97//6dixYwWe9x0A7OKMR8oDwI0gIyPDGDVqlNGkSRMjICDA8Pf3N5o0aWK8/fbbNutdbQq21157zWa93KmC/jnFUu50RVu3bs2zfnR0tBEUFGT4+PgYNWvWNAYPHmxs27bNZt/+/v7G77//bnTp0sXw8/MzwsLCjHHjxhk5OTk2/ekfU7AZxpXpiIYPH25Uq1bN8PT0NMLDw41OnToZ7777rr3DZVgsFmP+/PlGu3btjICAAEOSIcno3LlznmnJrmf79u1G//79jcqVKxuenp5G+fLljU6dOhnvv/++zXFduHDB+M9//mNdr3bt2sZrr71mM+1X7rH/cyo9ez6njh07Gg0bNjS2bdtmtGnTxvDx8TEiIiKMmTNn5hmDSZMmGREREYa3t7fRrFkzY+XKlQX+jvx92T+nYPP398+zbu4UZH+XkpJi9O/f3wgICDCCgoKMwYMHGz/88IMhyVi8eHGePvI79n9OfbZkyRKjWbNmhre3t1GhQgVjwIABxh9//JFn+z179hi9evUygoODDR8fH6Nu3brGiy++aF1+7tw5IzY21qhYsaJRrlw5Izo62ti3b58RERFhMzWavVOw/X2qQcO4+njlfo5/l5ycbK3Jy8vLaNSoUb5T1f3111/GwIEDjcDAQCMoKMgYOHCgddq0f67/+++/G4MGDTLCw8MNT09Po0qVKsbdd99tLF269KrHeOjQIWPIkCFGzZo1DR8fH6NChQrG7bffbqxZs8am76efftowmUzG3r17rzk2f5ffFGxPPPGEIcn4/fffr7rd+PHjDUnGrl27jOTkZGPKlClGx44djUqVKhkeHh5G+fLljTvuuMPmuAzj6r/TDMMwXn/9dUOScffdd9tMRflP9k7BdrWp66KioozAwEBj8ODBdh1vrvPnzxsvvfSS0axZM6NcuXKGl5eXUa1aNaNPnz7G559/ft36AKAwTIbBtToAcKMaPHiwli5dWiJnae2VlZWlHj16KCkpSZ9//rm6du3q7JIKLSoqSmfOnLnuPdeu6tNPP1WvXr30/fffq127ds4uB0XQqlUrRURE5HnyOgCg9OCedABAsfD09NSyZcvUtGlT3XfffUWaBxsFd+nSJZv3ufdOBwYGqnnz5k6qCo6QlpamXbt26aWXXnJ2KQCAYsQ96QCAq0pJSbnmQ728vLxUoUKFqy739/e3TnmEkvHEE0/o0qVLatOmjTIyMrR8+XJt3LhRkyZNsmvKN7iewMDAYn1oJQDANRDSAQBXdeutt+ro0aNXXd6xY0etX7++5ArCdd1xxx2aNm2aVq5cqcuXL6tWrVqaMWPGVR+YBQAAXAv3pAMAruqHH37Ic/n035UvX946/RgAAACKjpAOAAAAAICL4MFxAAAAAAC4iDJ3T7rFYtHJkycVEBAgk8nk7HIAAAAAAKWcYRi6cOGCKleuLDe3a58rL3Mh/eTJk6pWrZqzywAAAAAAlDHHjx9X1apVr7lOmQvpAQEBkq4MTmBgoJOrAQAAAACUdmlpaapWrZo1j15LmQvpuZe4BwYGEtIBAAAAACWmILdc8+A4AAAAAABcBCEdAAAAAAAXQUgHAAAAAMBFlLl70gEAAACgLDMMQ9nZ2crJyXF2KaWKp6en3N3di9wPIR0AAAAAyojMzEz9+eefMpvNzi6l1DGZTKpatarKlStXpH4I6QAAAABQBlgsFh0+fFju7u6qXLmyvLy8CvS0cVyfYRhKSUnRH3/8odq1axfpjDohHQAAAADKgMzMTFksFlWrVk1+fn7OLqfUCQkJ0ZEjR5SVlVWkkM6D4wAAAACgDHFzIwYWB0ddlcCnAwAAAACAi+BydwAAAAAow1JTU0v0QXJ+fn4KCgoqsf3daAjpAAAAAFBGpaamaubMRGVnZ5XYPj08PBUXN9yuoB4VFaWmTZsqISEh3+WRkZEaMWKERowY4ZginYiQDgAAAABllNlsVnZ2ljyad5apXIVi359x8ayyd6yR2WzmbPpVENIBAAAAoIwzlasgt+CQYt+Ppdj3cOPjwXEAAAAAgBvG6dOn1aNHD/n6+urmm2/WwoUL86xz/vx5PfroowoLC5OPj49uueUWrVy50rr8hx9+UFRUlPz8/FS+fHlFR0fr3LlzJXkYV8WZdAAAAADADWPw4ME6efKk1q1bJ09PTz355JM6ffq0dbnFYtFdd92lCxcu6MMPP1TNmjX166+/Wucu37lzpzp16qQhQ4bozTfflIeHh9atW6ecnBxnHZINQjoAAAAA4IZw4MABffnll9qyZYtuvfVWSdJ///tf1a9f37rOmjVrtGXLFu3du1d16tSRJNWoUcO6/NVXX1XLli319ttvW9saNmxYQkdwfVzuDgAAAAC4Iezdu1ceHh5q0aKFta1evXoKDg62vt+5c6eqVq1qDej/lHsm3VUR0gEAAAAApYavr2+Rljsbl7sDAACXlJqaKrPZXOR+/Pz8mOYHAEqJevXqKTs7W9u3b7de7r5//36dP3/euk7jxo31xx9/6MCBA/meTW/cuLGSkpI0YcKEkirbLoR0AADgclJTUzVzZqKys7OK3JeHh6fi4oYT1AHgGoyLZ0tkejTj4tkibV+3bl117dpVjz76qGbNmiUPDw+NGDHC5ux4x44d1aFDB/Xu3VvTp09XrVq1tG/fPplMJnXt2lVjxoxRo0aNNGzYMD322GPy8vLSunXrdN9996lixYpFPcQiI6QDAACXYzablZ2dJY/mnWUqV6HQ/RgXzyp7xxqZzWZCOgDkw8/PTx4ensresabE9unh4Sk/P79Cbz9v3jw9/PDD6tixo8LCwjRx4kS9+OKLNussW7ZMI0eOVL9+/ZSenq5atWppypQpkqQ6dero66+/1nPPPadWrVrJ19dXrVu3Vr9+/Yp0XI5CSAcAAC7LVK6C3IJDCr19SZwVAoAbWVBQkOLihjvk9qKCKsxtSOvXr7f+HB4ebjPnuSQNHDjQ5n2FChU0d+7cq/bXsWNH/fDDD3bVUFII6QAAAABQhgUFBXG1kQvh6e4AAAAAALgIQjoAAAAAAC6CkA4AAAAAgIsgpAMAAAAA4CII6QAAAAAAuAhCOgAAAAAALoKQDgAAAACAi2CedAAAAAAow1JTU2U2m0tsf35+fszLfg2EdAAAAAAoo1JTU5U4c6aysrNLbJ+eHh4aHhdnV1CPiopS06ZNlZCQoMjISI0YMUIjRowoviKdiJAOAAAAAGWU2WxWVna2etWorxBf/2LfX8qldK04tFdms5mz6VdBSAcAAACAMi7E11+V/AOcXQbEg+MAAAAAADew6dOnq1GjRvL391e1atU0bNgwXbx40bp8/vz5Cg4O1sqVK1W3bl35+fmpT58+MpvNev/99xUZGany5cvrySefVE5OjnW7BQsWqGXLlgoICFB4eLj69++v06dPF/vxENIBAAAAADcsNzc3vfXWW/rll1/0/vvva+3atXrmmWds1jGbzXrrrbe0ePFirV69WuvXr1evXr20atUqrVq1SgsWLNA777yjpUuXWrfJysrSyy+/rF27dunTTz/VkSNHNHjw4GI/Hi53BwAAAADcsP7+ALnIyEhNnDhRjz32mN5++21re1ZWlmbNmqWaNWtKkvr06aMFCxYoOTlZ5cqVU4MGDXT77bdr3bp16tu3ryRpyJAh1u1r1Kiht956S7feeqsuXryocuXKFdvxcCYdAAAAAHDDWrNmjTp16qQqVaooICBAAwcO1F9//WUzrZyfn581oEtSWFiYIiMjbcJ2WFiYzeXs27dvV48ePVS9enUFBASoY8eOkqRjx44V6/EQ0gEAAAAAN6QjR47o7rvvVuPGjbVs2TJt375diYmJkqTMzEzrep6enjbbmUymfNssFoskKT09XdHR0QoMDNTChQu1detWrVixIk+/xcHpIT0xMVGRkZHy8fFR69attWXLlmuuf/78eQ0fPlyVKlWSt7e36tSpo1WrVpVQtQAAAAAAV7F9+3ZZLBZNmzZN//rXv1SnTh2dPHmyyP3u27dPf/31l6ZMmaL27durXr16JfLQOMnJ96QvWbJE8fHxmj17tlq3bq2EhARFR0dr//79Cg0NzbN+Zmam7rzzToWGhmrp0qWqUqWKjh49quDg4JIvHgAAAABKiZRL6TfkfmrVqqWsrCzNmDFDPXr00A8//KDZs2cXud/q1avLy8tLM2bM0GOPPaY9e/bo5ZdfdkDF1+fUkD59+nQNHTpUsbGxkqTZs2friy++0Ny5czV69Og868+dO1dnz57Vxo0brZcmREZGlmTJAAAAAFBq+Pn5ydPDQysO7S2xfXp6eMjPz88hfTVp0kTTp0/X1KlTNWbMGHXo0EGTJ0/WoEGDitRvSEiI5s+fr+eee05vvfWWmjdvrtdff1333HOPQ+q+FpNhGEax7yUfmZmZ8vPz09KlS9WzZ09re0xMjM6fP6/PPvsszzbdunVThQoV5Ofnp88++0whISHq37+/nn32Wbm7u+e7n4yMDGVkZFjfp6WlqVq1ajpz5owCAwMdflwAAKDokpOTNW/ePHm26yW3oIqF7seSekZZP6xQbGyswsLCHFghANx4Ll++rOPHj1tvN86Vmppq85C14ubn56egoKAS219JuXz5so4cOaJq1arZjK90JYdWrFhRqamp182hTjuTfubMGeXk5OT5H2ZYWJj27duX7zaHDh3S2rVrNWDAAK1atUoHDx7UsGHDlJWVpXHjxuW7zeTJkzVhwoQ87V9//bXD/noDAAAcr3HjxtKF36+8itaRtm/f7piiAOAG5uHhofDwcF28eNHm4Wcmk0n+/v4lWktaWlqJ7q8kZGZm6tKlS/r222+VnZ1ts8yeP4LcUPOkWywWhYaG6t1335W7u7tatGihEydO6LXXXrtqSB8zZozi4+Ot73PPpHfp0oUz6QAAuCjOpAOA4+WeSS9XrlyeM70ousuXL8vX11cdOnTI90x6QTktpFesWFHu7u5KTk62aU9OTlZ4eHi+21SqVEmenp42l7bXr19fp06dUmZmpry8vPJs4+3tLW9v7zztnp6eeR65DwAAXIO7u7ssFossJnfJrfD/XLGYrvRz7ty5q94aV1Cl9fJMAGVHTk6OTCaT3Nzc5Obm9Im+Sh03Nzfr1G7/zJr2ZE+nhXQvLy+1aNFCSUlJ1nvSLRaLkpKSFBcXl+827dq106JFi2SxWKxfqgMHDqhSpUr5BnQAAFC2GZfTZZK0fPnyIvfl6eGh4XFxBHUAQLFy6uXu8fHxiomJUcuWLdWqVSslJCQoPT3d+rT3QYMGqUqVKpo8ebIk6fHHH9fMmTP11FNP6YknntBvv/2mSZMm6cknn3TmYQAAAFeVlSlDUq8a9RXiW/j7LVMupWvFob0ym82EdAA3PCc9O7zUc9S4OjWk9+3bVykpKRo7dqxOnTqlpk2bavXq1dZ7xo4dO2ZzGUa1atX01Vdf6T//+Y8aN26sKlWq6KmnntKzzz7rrEMAAAA3gBBff1XyD3B2GQDgVLmXXJvNZvn6+jq5mtIn92F8Rb29yukPjouLi7vq5e3r16/P09amTRtt3ry5mKsCAAAAgNLF3d1dwcHBOn36tKQrz9owmUxOrqp0sFgsSklJkZ+fnzw8ihaznR7SAQAAAAAlI/ch3blBHY7j5uam6tWrF/kPH4R0AAAAACgjTCaTKlWqpNDQUGVlZTm7nFLFy8vLIU/NJ6QDAAAAQBnj7u5e5HunUTyYHA8AAAAAABdBSAcAAAAAwEUQ0gEAAAAAcBGEdAAAAAAAXAQhHQAAAAAAF0FIBwAAAADARRDSAQAAAABwEYR0AAAAAABcBCEdAAAAAAAX4eHsAgAAAAAAZU9qaqrMZnOR+/Hz81NQUJADKnINhHQAAAAAQIlKTU1V4syZysrOLnJfnh4eGh4XV2qCOiEdAAAAAFCizGazsrKz1atGfYX4+he6n5RL6VpxaK/MZjMhHQAAAACAogjx9Vcl/wBnl+FSeHAcAAAAAAAugpAOAAAAAICLIKQDAAAAAOAiCOkAAAAAALgIHhwHAAAAACgwR8xvfubMGQdVU/oQ0gEAAAAABZKamqqZMxOVnZ3l7FJKLUI6AAAAAKBAzGazsrOz5NG8s0zlKhS6H8vpI8rZt8WBlZUehHQAAAAAgF1M5SrILTik0NsbF845sJrShQfHAQAAAADgIgjpAAAAAAC4CEI6AAAAAAAugpAOAAAAAICL4MFxAEqEI+bTlCQ/Pz8FBQU5oCIAAADA9RDSARQ7R86n6eHhqbi44QR1AAAAlEqEdADFzlHzaRoXzyp7xxqZzWZCOgAAAEolQjqAElPU+TQtDqwFAAAAcEU8OA4AAAAAABdBSAcAAAAAwEUQ0gEAAAAAcBGEdAAAAAAAXAQhHQAAAAAAF0FIBwAAAADARRDSAQAAAABwEYR0AAAAAABcBCEdAAAAAAAXQUgHAAAAAMBFENIBAAAAAHARhHQAAAAAAFwEIR0AAAAAABdBSAcAAAAAwEUQ0gEAAAAAcBGEdAAAAAAAXAQhHQAAAAAAF0FIBwAAAADARRDSAQAAAABwEYR0AAAAAABchIezC8C1paamymw2F7kfPz8/BQUFOaAiAAAAAEBxIaS7sNTUVM2cmajs7Kwi9+Xh4am4uOEEdQAAAABwYYR0F2Y2m5WdnSWP5p1lKleh0P0YF88qe8camc1mQjoAAAAAuDBC+g3AVK6C3IJDCr29xYG1AAAAAACKDw+OAwAAAADARRDSAQAAAABwEYR0AAAAAABchEuE9MTEREVGRsrHx0etW7fWli1brrru/PnzZTKZbF4+Pj4lWC0AAAAAAMXD6SF9yZIlio+P17hx47Rjxw41adJE0dHROn369FW3CQwM1J9//ml9HT16tAQrBgAAAACgeDj96e7Tp0/X0KFDFRsbK0maPXu2vvjiC82dO1ejR4/OdxuTyaTw8PAC9Z+RkaGMjAzr+7S0NElSVlaWsrKKPv94ccrJyZGbm5vcjBy5WbIL35FxpZ+cnByXP2aUTnyXAdjLUb83DMNy5feGDGUZRuHrkcHvHwAQv58Ly57aTIZRhBEposzMTPn5+Wnp0qXq2bOntT0mJkbnz5/XZ599lmeb+fPn6+GHH1aVKlVksVjUvHlzTZo0SQ0bNsx3H+PHj9eECRPytC9atEh+fn4OOxYAAAAAAPJjNpvVv39/paamKjAw8JrrOvVM+pkzZ5STk6OwsDCb9rCwMO3bty/fberWrau5c+eqcePGSk1N1euvv662bdvql19+UdWqVfOsP2bMGMXHx1vfp6WlqVq1aurSpct1B8fZkpOTNW/ePHm26yW3oIqF7seSekZZP6xQbGxsnrEGSgLfZQD2ctTvjZw/Dip79zrF1m+mML+AwtdjvqB5e3/i9w+AMo/fz4WTe0V3QTj9cnd7tWnTRm3atLG+b9u2rerXr6933nlHL7/8cp71vb295e3tnafd09NTnp6exVprUbm7u8tischicpfcCv9RWUxX+nF3d3f5Y0bpxHcZgL0c93vD7crvDZnkaTIVvh6ZSu3vn9TUVJnN5iL34+fnp6CgIAdUBMCV8fu5cOypzakhvWLFinJ3d1dycrJNe3JycoHvOff09FSzZs108ODB4igRAACg1EpNTdXMmYnKzi76fZweHp6KixtOUAeAInJqSPfy8lKLFi2UlJRkvSfdYrEoKSlJcXFxBeojJydHP//8s7p161aMlQIAAJQ+ZrNZ2dlZ8mjeWaZyFQrdj3HxrLJ3rJHZbCakA0AROf1y9/j4eMXExKhly5Zq1aqVEhISlJ6ebn3a+6BBg1SlShVNnjxZkvTSSy/pX//6l2rVqqXz58/rtdde09GjR/Xwww878zAAAABuWKZyFeQWHFLo7S0OrAUAyjqnh/S+ffsqJSVFY8eO1alTp9S0aVOtXr3aetP/sWPH5Ob2f9O5nzt3TkOHDtWpU6dUvnx5tWjRQhs3blSDBg2cdQgAAAAAADiE00O6JMXFxV318vb169fbvH/jjTf0xhtvlEBVAAAAAACULLfrrwIAAAAAAEoCIR0AAAAAABfhEpe7A4A9zpw5U+Q+mM8XAAAAroiQDuCGYVxOl0nS8uXLi9yXp4eHhsfFEdQBAADgUgjpAG4cWZkyJPWqUV8hvv6F7iblUrpWHNrLfL4AAABwOYR0ADecEF9/VfIPcHYZAAAAgMPx4DgAAAAAAFwEIR0AAAAAABdBSAcAAAAAwEXYFdINw9CxY8d0+fLl4qoHAAAAAIAyy+6QXqtWLR0/fry46gEAAAAAoMyyK6S7ubmpdu3a+uuvv4qrHgAAAAAAyiy770mfMmWKRo0apT179hRHPQAAAAAAlFl2z5M+aNAgmc1mNWnSRF5eXvL19bVZfvbsWYcVBwAAAABAWWJ3SE9ISCiGMgAAAAAAgN0hPSYmpjjqAAAAAACgzCvUPOm///67XnjhBfXr10+nT5+WJH355Zf65ZdfHFocAAAAAABlid0hfcOGDWrUqJF+/PFHLV++XBcvXpQk7dq1S+PGjXN4gQAAAAAAlBV2X+4+evRoTZw4UfHx8QoICLC233HHHZo5c6ZDiwMAAABKg9TUVJnN5iL34+fnp6CgIAdUBMBV2R3Sf/75Zy1atChPe2hoqM6cOeOQogAAAIDSIjU1VTNnJio7O6vIfXl4eCoubjhBHSjF7A7pwcHB+vPPP3XzzTfbtP/000+qUqWKwwoDAAAASgOz2azs7Cx5NO8sU7kKhe7HuHhW2TvWyGw2E9KBUszukP7AAw/o2Wef1SeffCKTySSLxaIffvhBI0eO1KBBg4qjRqBYcfkZAAAoCaZyFeQWHFLo7S0OrAWA67I7pE+aNEnDhw9XtWrVlJOTowYNGignJ0f9+/fXCy+8UBw1AsWGy88AAAAAuBK7Q7qXl5fmzJmjF198UXv27NHFixfVrFkz1a5duzjqA4oVl58BAAAAcCV2h/RDhw6pRo0aql69uqpXr14cNQEljsvPAAAAALgCu0N6rVq1VLVqVXXs2FFRUVHq2LGjatWqVRy1AQAAAABQprjZu8Hx48c1efJk+fr66tVXX1WdOnVUtWpVDRgwQO+9915x1AgAAAAAQJlgd0ivUqWKBgwYoHfffVf79+/X/v371blzZ3388cd69NFHi6NGAAAAAADKBLsvdzebzfr++++1fv16rV+/Xj/99JPq1aunuLg4RUVFFUOJAAAAAACUDXaH9ODgYJUvX14DBgzQ6NGj1b59e5UvX744agMAAAAAoEyxO6R369ZN33//vRYvXqxTp07p1KlTioqKUp06dYqjPgAAAAAAygy770n/9NNPdebMGa1evVpt2rTR119/rfbt21vvVQcAAAAAAIVj95n0XI0aNVJ2drYyMzN1+fJlffXVV1qyZIkWLlzoyPoAAAAAACgz7D6TPn36dN1zzz266aab1Lp1a3300UeqU6eOli1bppSUlOKoEQAAAACAMsHuM+kfffSROnbsqEceeUTt27dXUFBQcdQFAAAAAECZY3dI37p1a3HUAQAAAABAmWf35e6rV6/W999/b32fmJiopk2bqn///jp37pxDiwMAAAAAoCyxO6SPGjVKaWlpkqSff/5ZTz/9tLp166bDhw8rPj7e4QUCAAAAAFBW2H25++HDh9WgQQNJ0rJly3T33Xdr0qRJ2rFjh7p16+bwAgEAAAAAKCvsPpPu5eUls9ksSVqzZo26dOkiSapQoYL1DDsAAAAAALCf3WfSb7vtNsXHx6tdu3basmWLlixZIkk6cOCAqlat6vACAQAAAAAoK+w+kz5z5kx5eHho6dKlmjVrlqpUqSJJ+vLLL9W1a1eHFwgAAAAAQFlh95n06tWra+XKlXna33jjDYcUBAAAAABAWWV3SJcki8WigwcP6vTp07JYLDbLOnTo4JDCAAAAAAAoa+wO6Zs3b1b//v119OhRGYZhs8xkMiknJ8dhxQEAAAAAUJbYHdIfe+wxtWzZUl988YUqVaokk8lUHHUBAAAAAFDm2B3Sf/vtNy1dulS1atUqjnoAAAAAACiz7H66e+vWrXXw4MHiqAUAAAAAgDLN7jPpTzzxhJ5++mmdOnVKjRo1kqenp83yxo0bO6w4AAAAAADKErtDeu/evSVJQ4YMsbaZTCYZhsGD4wAAAAAAKAK7Q/rhw4eLow4AAAAAAMo8u0N6REREvu0Wi0WrVq266nIAAAAAAHBtdof0fzp48KDmzp2r+fPnKyUlRVlZWY6oCwAAAACAMsfup7tL0qVLl/TBBx+oQ4cOqlu3rjZu3KixY8fqjz/+cHR9AAAAAACUGXadSd+6davee+89LV68WDVr1tSAAQO0ceNGvf3222rQoEFx1QgAAAAAQJlQ4JDeuHFjpaWlqX///tq4caMaNmwoSRo9enSxFQcAAAAAQFlS4Mvd9+/frw4dOuj222/nrDkAAAAAAMWgwCH90KFDqlu3rh5//HFVrVpVI0eO1E8//SSTyVSc9QEAAAAAUGYUOKRXqVJFzz//vA4ePKgFCxbo1KlTateunbKzszV//nwdOHCg0EUkJiYqMjJSPj4+at26tbZs2VKg7RYvXiyTyaSePXsWet8AAAAAALiKQj3d/Y477tCHH36oP//8UzNnztTatWtVr149NW7c2O6+lixZovj4eI0bN047duxQkyZNFB0drdOnT19zuyNHjmjkyJFq3759YQ4BAAAAAACXU6iQnisoKEjDhg3Ttm3btGPHDkVFRdndx/Tp0zV06FDFxsaqQYMGmj17tvz8/DR37tyrbpOTk6MBAwZowoQJqlGjRhGOAAAAAAAA12HXFGzX0rRpU7311lt2bZOZmant27drzJgx1jY3Nzd17txZmzZtuup2L730kkJDQ/XQQw/pu+++u+Y+MjIylJGRYX2flpYmScrKylJWVpZd9Za0nJwcubm5yc3IkZslu/AdGVf6ycnJcfljLmmMcclw1DgbhuXKOMtQlmEUvh4ZfF6Ai+P3Rsng/4Mlg3FGacLv58KxpzaTYRRhRIro5MmTqlKlijZu3Kg2bdpY25955hlt2LBBP/74Y55tvv/+ez3wwAPauXOnKlasqMGDB+v8+fP69NNP893H+PHjNWHChDztixYtkp+fn8OOBQAAAACA/JjNZvXv31+pqakKDAy85roOO5NeEi5cuKCBAwdqzpw5qlixYoG2GTNmjOLj463v09LSVK1aNXXp0uW6g+NsycnJmjdvnjzb9ZJbUMGONz+W1DPK+mGFYmNjFRYW5sAKb3yMcclw1Djn/HFQ2bvXKbZ+M4X5BRS+HvMFzdv7E58X4ML4vVEy+P9gyWCcUZrw+7lwcq/oLginhvSKFSvK3d1dycnJNu3JyckKDw/Ps/7vv/+uI0eOqEePHtY2i8UiSfLw8ND+/ftVs2ZNm228vb3l7e2dpy9PT095eno64jCKjbu7uywWiywmd8mt8B+VxXSlH3d3d5c/5pLGGJcMx42z25VxlkmeRZj+0V0mPi/AxfF7o2Tw/8GSwTijNOH3c+HYU5vdD4774IMPbO7xzpWZmakPPvjArr68vLzUokULJSUlWdssFouSkpJsLn/PVa9ePf3888/auXOn9XXPPffo9ttv186dO1WtWjV7DwcAAAAAAJdhd0iPjY1VampqnvYLFy4oNjbW7gLi4+M1Z84cvf/++9q7d68ef/xxpaenW/saNGiQ9cFyPj4+uuWWW2xewcHBCggI0C233CIvLy+79w8AAAAAgKuw+/oEwzBkyudyhD/++ENBQUF2F9C3b1+lpKRo7NixOnXqlJo2barVq1db7yc4duyY3NyKNFMcAAAAgDIgNTVVZrO5yP34+fkVKtsAjlDgkN6sWTOZTCaZTCZ16tRJHh7/t2lOTo4OHz6srl27FqqIuLg4xcXF5bts/fr119x2/vz5hdonAAAAgNIjNTVVM2cmKju76NNweXh4Ki5uOEEdTlHgkN6zZ09J0s6dOxUdHa1y5cpZl3l5eSkyMlK9e/d2eIEAAAAAcD1ms1nZ2VnyaN5ZpnIVCt2PcfGssneskdlsJqTDKQoc0seNG6ecnBxFRkaqS5cuqlSpUnHWBQAAAAB2M5WrILfgkEJvb3FgLUBh2HWzt7u7ux599FFdvny5uOoBAAAAAKDMsvuJbLfccosOHTpUHLUAAAAAAFCm2R3SJ06cqJEjR2rlypX6888/lZaWZvMCAAAAAACFY/cUbN26dZMk3XPPPTZTseVOzZaTk+O46gAAAAAAKEPsDulr167Nd550AAAAAABQNHaH9KioqGIoAwAAAAAA2B3SO3TooKioKHXs2FHt2rWTj49PcdQFAAAAAECZY/eD47p06aLNmzfr3nvvVXBwsG677Ta98MIL+uabb2Q2m4ujRgAAAAAAygS7z6S/8MILkqTs7Gxt3bpVGzZs0Pr16/Xqq6/Kzc2NOdQBAAAAACgku0N6rkOHDunnn3/Wrl27tHv3bgUEBKhDhw6OrA0AAAAAgDLF7pDev39/bdiwQRkZGerQoYM6duyo0aNHq3Hjxjz1HWXemTNnityHn5+fgoKCHFANAAAAgBuN3SF98eLFqlixoh5++GHdcccduu222+Tn51cctQE3DONyukySli9fXuS+PD08NDwujqAOAAAAlEF2h/S//vpL3333ndavX68xY8Zo7969atq0qaKiohQVFaUuXboUR52Aa8vKlCGpV436CvH1L3Q3KZfSteLQXpnNZkI6AAAAUAbZHdLLly+ve+65R/fcc48k6eDBg5o4caJee+01TZ06VTk5OQ4vErhRhPj6q5J/gLPLAAAAAHCDKtSZ9Nwnuq9fv16//vqrgoOD1aNHD3Xs2LE4agQAAAAAoEywO6SHhoaqYsWKat++vYYOHaqoqCg1atSoOGoDAAAAAKBMsTuk7969Ww0bNiyOWgAAAAAAKNPc7N2AgA4AAAAAQPGwO6QDAAAAAIDiQUgHAAAAAMBF2H1POgAAAJCfM2fOFLkPPz8/BQUFOaAaALgx2R3SDx06pBo1ahRHLQAAALgBGZfTZZK0fPnyIvfl6eGh4XFxBHUAZZbdIb1WrVrq2LGjHnroIfXp00c+Pj7FURcAAABuFFmZMiT1qlFfIb7+he4m5VK6VhzaK7PZTEgHUGbZHdJ37NihefPmKT4+XnFxcerbt68eeughtWrVqjjqAwDYITU1VWazucj9cLkpgMII8fVXJf8AZ5cBADc0u0N606ZN9eabb2ratGn63//+p/nz5+u2225TnTp1NGTIEA0cOFAhISHFUSsA4BpSU1M1c2aisrOzityXh4en4uKGE9QBAABKWKEfHOfh4aF///vf6t69u95++22NGTNGI0eO1HPPPaf7779fU6dOVaVKlRxZKwDgGsxms7Kzs+TRvLNM5SoUuh/j4lll71jD5aYAAABOUOiQvm3bNs2dO1eLFy+Wv7+/Ro4cqYceekh//PGHJkyYoHvvvVdbtmxxZK0AgAIwlasgt+DCX9FkcWAtAAAAsI/dIX369OmaN2+e9u/fr27duumDDz5Qt27d5OZ2Zcr1m2++WfPnz1dkZKSjawUAAAAAoFSzO6TPmjVLQ4YM0eDBg696OXtoaKj++9//Frk4AAAAAADKErtD+jfffKPq1atbz5znMgxDx48fV/Xq1eXl5aWYmBiHFQkAAAAAQFngdv1VbNWsWVNnzpzJ03727FndfPPNDikKAAAAAICyyO6QbhhGvu0XL16Uj49PkQsCAAAAAKCsKvDl7vHx8ZIkk8mksWPHys/Pz7osJydHP/74o5o2berwAgEAAAAAKCsKHNJ/+uknSVfOpP/888/y8vKyLvPy8lKTJk00cuRIx1cIAAAAAEAZUeCQvm7dOklSbGys3nzzTQUGBhZbUQAAAAAAlEV2P9193rx5xVEHAAAAAABlXoFC+r///W/Nnz9fgYGB+ve//33NdZcvX+6QwgAAAAAAKGsKFNKDgoJkMpmsPwMAAAAAAMcrUEjPvcTdMAxNmDBBISEh8vX1LdbCAAAAAAAoa+yaJ90wDNWqVUt//PFHcdUDAAAAAECZZVdId3NzU+3atfXXX38VVz0AAAAAAJRZdoV0SZoyZYpGjRqlPXv2FEc9AAAAAACUWXZPwTZo0CCZzWY1adJEXl5eee5NP3v2rMOKAwAAAACgLLE7pCckJBRDGQAAAAAAwO6QHhMTUxx1AAAAAABQ5tkd0v/u8uXLyszMtGkLDAwsUkEAAAAAAJRVdj84Lj09XXFxcQoNDZW/v7/Kly9v8wIAAAAAAIVjd0h/5plntHbtWs2aNUve3t567733NGHCBFWuXFkffPBBcdQIAAAAAECZYPfl7p9//rk++OADRUVFKTY2Vu3bt1etWrUUERGhhQsXasCAAcVRJwAAAAAApZ7dIf3s2bOqUaOGpCv3n+dOuXbbbbfp8ccfd2x1AACnOXPmTJH78PPzU1BQkAOqAQAAKBvsDuk1atTQ4cOHVb16ddWrV08ff/yxWrVqpc8//1zBwcHFUCIAoCQZl9NlkrR8+fIi9+Xp4aHhcXEEdQAAgAKyO6THxsZq165d6tixo0aPHq0ePXpo5syZysrK0vTp04ujRgBAScrKlCGpV436CvH1L3Q3KZfSteLQXpnNZkI6AABAAdkd0v/zn/9Yf+7cubP27dun7du3q1atWmrcuLFDiwMAOE+Ir78q+Qc4uwwAAIAypUjzpEtSRESEIiIiHFELAAAAAABlWoFC+ltvvVXgDp988slCFwMAAAAAQFlWoJD+xhtvFKgzk8lESAcAAABww2OWEzhLgUL64cOHi7WIxMREvfbaazp16pSaNGmiGTNmqFWrVvmuu3z5ck2aNEkHDx5UVlaWateuraeffloDBw4s1hoBAAAAlH7McgJnK/I96UW1ZMkSxcfHa/bs2WrdurUSEhIUHR2t/fv3KzQ0NM/6FSpU0PPPP6969erJy8tLK1euVGxsrEJDQxUdHe2EIwAAAABQajDLCZysQCE9Pj5eL7/8svz9/RUfH3/Nde2dhm369OkaOnSoYmNjJUmzZ8/WF198oblz52r06NF51o+KirJ5/9RTT+n999/X999/T0gHAAAA4BDMcgJnKVBI/+mnn5SVlWX9+WpMJpNdO8/MzNT27ds1ZswYa5ubm5s6d+6sTZs2XXd7wzC0du1a7d+/X1OnTs13nYyMDGVkZFjfp6WlSZKysrKsx+SqcnJy5ObmJjcjR26W7MJ3ZFzpJycnx+WPuaQ5aowNw3JljGUoyzAKX4+MUvlZMc4lg3FGacL3uWQwziWDf9OVDL7PJYNxLhx7ajMZRhFGpIhOnjypKlWqaOPGjWrTpo21/ZlnntGGDRv0448/5rtdamqqqlSpooyMDLm7u+vtt9/WkCFD8l13/PjxmjBhQp72RYsWyc/PzzEHAgAAAADAVZjNZvXv31+pqakKDAy85rpOvye9MAICArRz505dvHhRSUlJio+PV40aNfJcCi9JY8aMsblEPy0tTdWqVVOXLl2uOzjOlpycrHnz5smzXS+5BVUsdD+W1DPK+mGFYmNjFRYW5sAKb3yOGuOcPw4qe/c6xdZvpjC/wl8WlWy+oHl7fyp1nxXjXDIYZ5QmfJ9LBuNcMhw2zsnHlLP9KzniDJunh4eGPvKIy/972B58n0sG41w4uVd0F4TdIf3y5cuaMWOG1q1bp9OnT8tisdgs37FjR4H7qlixotzd3ZWcnGzTnpycrPDw8Ktu5+bmplq1akmSmjZtqr1792ry5Mn5hnRvb295e3vnaff09JSnp2eBa3UGd3d3WSwWWUzuklvh/55iMV3px93d3eWPuaQ5bozdroyxTPK087YPm3pkKpWfFeNcMhhnlCZ8n0sG41wyHDbO2dnKsVgc9kCzzMxMxjkffJ+vjXEuHHtqs3tUH3roIX399dfq06ePWrVqZfd96H/n5eWlFi1aKCkpST179pQkWSwWJSUlKS4ursD9WCwWm/vOAQAAgNKKB5oBpZvdIX3lypVatWqV2rVr55AC4uPjFRMTo5YtW6pVq1ZKSEhQenq69WnvgwYNUpUqVTR58mRJ0uTJk9WyZUvVrFlTGRkZWrVqlRYsWKBZs2Y5pB4AAAAAAJzF7pBepUoVBQQ47i93ffv2VUpKisaOHatTp06padOmWr16tfV+gmPHjsnNzc26fnp6uoYNG6Y//vhDvr6+qlevnj788EP17dvXYTUBAAAAAOAMdof0adOm6dlnn9Xs2bMVERHhkCLi4uKuenn7+vXrbd5PnDhREydOdMh+AQAAAABwJXaH9JYtW+ry5cuqUaOG/Pz88twAf/bsWYcVBwAAAABAWWJ3SO/Xr59OnDihSZMmKSwsrEgPjgMAAAAAAP/H7pC+ceNGbdq0SU2aNCmOegAAAAAAKLPcrr+KrXr16unSpUvFUQsAAAAAAGWa3SF9ypQpevrpp7V+/Xr99ddfSktLs3kBAAAAAIDCsfty965du0qSOnXqZNNuGIZMJpNycnIcUxkAAAAAAGWM3SF93bp1xVEHAAAAAABlnt0hvWPHjsVRBwAAAAAAZV6BQvru3bt1yy23yM3NTbt3777muo0bN3ZIYQAAAAAAlDUFCulNmzbVqVOnFBoaqqZNm8pkMskwjDzrcU86AAAAAACFV6CQfvjwYYWEhFh/BgAAAAAAjlegkB4REZHvzwAAAAAAwHEKPE/6gQMHtGXLFpu2pKQk3X777WrVqpUmTZrk8OIAAAAAAChLChzSn332Wa1cudL6/vDhw+rRo4e8vLzUpk0bTZ48WQkJCcVRIwAAAAAAZUKBp2Dbtm2bnnnmGev7hQsXqk6dOvrqq68kXXmq+4wZMzRixAiHFwkAAAAAQFlQ4DPpZ86cUdWqVa3v161bpx49eljfR0VF6ciRIw4tDgAAAACAsqTAIb1ChQr6888/JUkWi0Xbtm3Tv/71L+vyzMzMfKdlAwAAAAAABVPgkB4VFaWXX35Zx48fV0JCgiwWi6KioqzLf/31V0VGRhZDiQAAAAAAlA0Fvif9lVde0Z133qmIiAi5u7vrrbfekr+/v3X5ggULdMcddxRLkQAAAAAAlAUFDumRkZHau3evfvnlF4WEhKhy5co2yydMmGBzzzoAAAAAALBPgUO6JHl4eKhJkyb5LrtaOwAAAAAAKJgC35MOAAAAAACKFyEdAAAAAAAXQUgHAAAAAMBFENIBAAAAAHARhQrp3333nR588EG1adNGJ06ckHRlCrbvv//eocUBAAAAAFCW2B3Sly1bpujoaPn6+uqnn35SRkaGJCk1NVWTJk1yeIEAAAAAAJQVdof0iRMnavbs2ZozZ448PT2t7e3atdOOHTscWhwAAAAAAGWJ3SF9//796tChQ572oKAgnT9/3hE1AQAAAABQJtkd0sPDw3Xw4ME87d9//71q1KjhkKIAAAAAACiL7A7pQ4cO1VNPPaUff/xRJpNJJ0+e1MKFCzVy5Eg9/vjjxVEjAAAAAABlgoe9G4wePVoWi0WdOnWS2WxWhw4d5O3trZEjR+qJJ54ojhoBAAAAACgT7A7pJpNJzz//vEaNGqWDBw/q4sWLatCggcqVK1cc9QEAAAAAUGbYHdJzeXl5qUGDBo6sBQAAAACAMs3ukJ6enq4pU6YoKSlJp0+flsVisVl+6NAhhxUHAAAAAEBZYndIf/jhh7VhwwYNHDhQlSpVkslkKo66AAAAAAAoc+wO6V9++aW++OILtWvXrjjqAQAAAACgzLJ7Crby5curQoUKxVELAAAAAABlmt0h/eWXX9bYsWNlNpuLox4AAAAAAMosuy93nzZtmn7//XeFhYUpMjJSnp6eNst37NjhsOIAAAAAAChL7A7pPXv2LIYyAAAAAACA3SF93LhxxVEHAAAAAABlnt33pAMAAAAAgOJRoDPpFSpU0IEDB1SxYkWVL1/+mnOjnz171mHFAQAAAABQlhQopL/xxhsKCAiw/nytkA4AAAAAAAqnQCE9JibG+vPgwYOLqxYAAAAAAMo0u+9J37Fjh37++Wfr+88++0w9e/bUc889p8zMTIcWBwAAAABAWWJ3SH/00Ud14MABSdKhQ4fUt29f+fn56ZNPPtEzzzzj8AIBAAAAACgr7A7pBw4cUNOmTSVJn3zyiTp27KhFixZp/vz5WrZsmaPrAwAAAACgzLA7pBuGIYvFIklas2aNunXrJkmqVq2azpw549jqAAAAAAAoQ+wO6S1bttTEiRO1YMECbdiwQd27d5ckHT58WGFhYQ4vEAAAAACAssLukJ6QkKAdO3YoLi5Ozz//vGrVqiVJWrp0qdq2bevwAgEAAAAAKCsKNAXb3zVu3Njm6e65XnvtNbm7uzukKAAAAAAAyiK7Q3qu7du3a+/evZKkBg0aqHnz5g4rCgAAAACAssjukH769Gn17dtXGzZsUHBwsCTp/Pnzuv3227V48WKFhIQ4ukYAAAAAAMoEu+9Jf+KJJ3Tx4kX98ssvOnv2rM6ePas9e/YoLS1NTz75ZHHUCAAAAABAmWD3mfTVq1drzZo1ql+/vrWtQYMGSkxMVJcuXRxaHAAAAAAAZYndZ9ItFos8PT3ztHt6elrnTwcAAAAAAPazO6Tfcccdeuqpp3Ty5Elr24kTJ/Sf//xHnTp1KlQRiYmJioyMlI+Pj1q3bq0tW7Zcdd05c+aoffv2Kl++vMqXL6/OnTtfc30AAAAAAG4Udof0mTNnKi0tTZGRkapZs6Zq1qypm2++WWlpaZoxY4bdBSxZskTx8fEaN26cduzYoSZNmig6OlqnT5/Od/3169erX79+WrdunTZt2qRq1aqpS5cuOnHihN37BgAAAADAldh9T3q1atW0Y8cOrVmzRvv27ZMk1a9fX507dy5UAdOnT9fQoUMVGxsrSZo9e7a++OILzZ07V6NHj86z/sKFC23ev/fee1q2bJmSkpI0aNCgQtUAAAAAAIArKNQ86SaTSXfeeafuvPPOIu08MzNT27dv15gxY6xtbm5u6ty5szZt2lSgPsxms7KyslShQoV8l2dkZCgjI8P6Pi0tTZKUlZWlrKysIlRf/HJycuTm5iY3I0duluzCd2Rc6ScnJ8flj7mkOWqMDcNyZYxlKMswCl+PjFL5WTHOJYNxRmnC97lkMM4lg3EuGYxzyWCcC8ee2kyGUbARWbt2reLi4rR582YFBgbaLEtNTVXbtm01e/ZstW/fvsA7P3nypKpUqaKNGzeqTZs21vZnnnlGGzZs0I8//njdPoYNG6avvvpKv/zyi3x8fPIsHz9+vCZMmJCnfdGiRfLz8ytwrQAAAAAAFIbZbFb//v2VmpqaJ0//U4HPpCckJGjo0KH5dhgUFKRHH31U06dPtyukF9WUKVO0ePFirV+/Pt+ALkljxoxRfHy89X1aWpr1PvbrDY6zJScna968efJs10tuQRUL3Y8l9Yyyflih2NhYhYWFObDCG5+jxjjnj4PK3r1OsfWbKcwvoPD1mC9o3t6fSt1nxTiXDMYZpQnf55LBOJcMxrlkMM4lg3EunNwruguiwCF9165dmjp16lWXd+nSRa+//nqBdyxJFStWlLu7u5KTk23ak5OTFR4efs1tX3/9dU2ZMkVr1qxR48aNr7qet7e3vL2987R7enrmO5WcK3F3d5fFYpHF5C65FerOBEmSxXSlH3d3d5c/5pLmuDF2uzLGMsnTZCp8PTKVys+KcS4ZjDNKE77PJYNxLhmMc8lgnEsG41w49tRW4Ke7JycnX7NjDw8PpaSkFHjHkuTl5aUWLVooKSnJ2maxWJSUlGRz+fs/vfrqq3r55Ze1evVqtWzZ0q59AgAAAADgqgoc0qtUqaI9e/Zcdfnu3btVqVIluwuIj4/XnDlz9P7772vv3r16/PHHlZ6ebn3a+6BBg2weLDd16lS9+OKLmjt3riIjI3Xq1CmdOnVKFy9etHvfAAAAAAC4kgJfn9CtWze9+OKL6tq1a577vy9duqRx48bp7rvvtruAvn37KiUlRWPHjtWpU6fUtGlTrV692no/wbFjx+Tm9n9/S5g1a5YyMzPVp08fm37GjRun8ePH271/AAAAAABcRYFD+gsvvKDly5erTp06iouLU926dSVJ+/btU2JionJycvT8888Xqoi4uDjFxcXlu2z9+vU2748cOVKofQAAAAAA4OoKHNLDwsK0ceNGPf744xozZoxyZ24zmUyKjo5WYmKiSz9NDwAAAAAAV2fX4/giIiK0atUqnTt3TgcPHpRhGKpdu7bKly9fXPUBAAAAAFBmFOqZ+eXLl9ett97q6FoAAAAAACjTCvx0dwAAAAAAULwKP/s8AAAoktTUVJnN5iL34+fnp6CgIAdUBAAAnI2QDgCAE6Smpipx5kxlZWcXuS9PDw8Nj4sjqAMAUAoQ0gEAcAKz2ays7Gz1qlFfIb7+he4n5VK6VhzaK7PZTEgHAKAUIKQDAOBEIb7+quQf4OwyAACAi+DBcQAAAAAAuAhCOgAAAAAALoKQDgAAAACAiyCkAwAAAADgIgjpAAAAAAC4CEI6AAAAAAAugpAOAAAAAICLIKQDAAAAAOAiCOkAAAAAALgID2cXgJJz5syZIvfh5+enoKAgB1QDAAAAAPgnQnoZYFxOl0nS8uXLi9yXp4eHhsfFEdQBAAAAoBgQ0suCrEwZknrVqK8QX/9Cd5NyKV0rDu2V2WwmpAMAAABAMSCklyEhvv6q5B/g7DIAAAAAAFfBg+MAAAAAAHARhHQAAAAAAFwEIR0AAAAAABdBSAcAAAAAwEUQ0gEAAAAAcBGEdAAAAAAAXAQhHQAAAAAAF0FIBwAAAADARRDSAQAAAABwEYR0AAAAAABcBCEdAAAAAAAXQUgHAAAAAMBFENIBAAAAAHARhHQAAAAAAFwEIR0AAAAAABdBSAcAAAAAwEUQ0gEAAAAAcBGEdAAAAAAAXAQhHQAAAAAAF0FIBwAAAADARRDSAQAAAABwEYR0AAAAAABcBCEdAAAAAAAXQUgHAAAAAMBFENIBAAAAAHARhHQAAAAAAFwEIR0AAAAAABdBSAcAAAAAwEUQ0gEAAAAAcBGEdAAAAAAAXAQhHQAAAAAAF0FIBwAAAADARRDSAQAAAABwEYR0AAAAAABcBCEdAAAAAAAXQUgHAAAAAMBFENIBAAAAAHARhHQAAAAAAFyE00N6YmKiIiMj5ePjo9atW2vLli1XXfeXX35R7969FRkZKZPJpISEhJIrFAAAAACAYubUkL5kyRLFx8dr3Lhx2rFjh5o0aaLo6GidPn063/XNZrNq1KihKVOmKDw8vISrBQAAAACgeHk4c+fTp0/X0KFDFRsbK0maPXu2vvjiC82dO1ejR4/Os/6tt96qW2+9VZLyXZ6fjIwMZWRkWN+npaVJkrKyspSVlVXUQyhWOTk5cnNzk5uRIzdLdqH7MQyL3NzclCNDWYZR+HpkXOknJ8flx66gGOOSwTiXDMb5xpL7eTHO+eP7XDIY55LBOJcMxrlkMM6FY09tJsMowogUQWZmpvz8/LR06VL17NnT2h4TE6Pz58/rs88+u+b2kZGRGjFihEaMGHHN9caPH68JEybkaV+0aJH8/PwKUzoAAAAAAAVmNpvVv39/paamKjAw8JrrOu1M+pkzZ5STk6OwsDCb9rCwMO3bt89h+xkzZozi4+Ot79PS0lStWjV16dLluoPjbMnJyZo3b5482/WSW1DFQveT88dBZe9ep9j6zRTmF1D4eswXNG/vT4qNjc3zud2oGOOSwTiXDMb5xpL7eTHO+eP7XDIY55LBOJcMxrlkMM6Fk3tFd0E49XL3kuDt7S1vb+887Z6envL09HRCRQXn7u4ui8Uii8ldciv8R2UxuclischdJnmaTIWvR6Yr/bi7u/zYFRRjXDIY55LBON9Ycj8vxjl/fJ9LBuNcMhjnksE4lwzGuXDsqc1pD46rWLGi3N3dlZycbNOenJzMQ+EAAAAAAGWS00K6l5eXWrRooaSkJGubxWJRUlKS2rRp46yyAAAAAABwGqde7h4fH6+YmBi1bNlSrVq1UkJCgtLT061Pex80aJCqVKmiyZMnS7rysLlff/3V+vOJEye0c+dOlStXTrVq1XLacQAAAAAA4AhODel9+/ZVSkqKxo4dq1OnTqlp06ZavXq19Yb/Y8eOyc3t/072nzx5Us2aNbO+f/311/X666+rY8eOWr9+fUmXDwAAAACAQzn9wXFxcXGKi4vLd9k/g3dkZKScNGMcAAAAAADFzmn3pAMAAAAAAFuEdAAAAAAAXAQhHQAAAAAAF0FIBwAAAADARRDSAQAAAABwEYR0AAAAAABcBCEdAAAAAAAXQUgHAAAAAMBFENIBAAAAAHARhHQAAAAAAFwEIR0AAAAAABdBSAcAAAAAwEUQ0gEAAAAAcBGEdAAAAAAAXAQhHQAAAAAAF0FIBwAAAADARRDSAQAAAABwEYR0AAAAAABcBCEdAAAAAAAXQUgHAAAAAMBFENIBAAAAAHARhHQAAAAAAFwEIR0AAAAAABdBSAcAAAAAwEUQ0gEAAAAAcBGEdAAAAAAAXAQhHQAAAAAAF+Hh7AIAALjRpKamymw2F6mPM2fOOKgaAABQmhDSAQCwQ2pqqmbOTFR2dpazSwEAAKUQIR0AADuYzWZlZ2fJo3lnmcpVKHQ/ltNHlLNviwMrAwAApQEhHQCAQjCVqyC34JBCb29cOOfAagAAQGnBg+MAAAAAAHARhHQAAAAAAFwEIR0AAAAAABdBSAcAAAAAwEUQ0gEAAAAAcBGEdAAAAAAAXAQhHQAAAAAAF0FIBwAAAADARRDSAQAAAABwEYR0AAAAAABcBCEdAAAAAAAXQUgHAAAAAMBFENIBAAAAAHARhHQAAAAAAFwEIR0AAAAAABdBSAcAAAAAwEUQ0gEAAAAAcBGEdAAAAAAAXAQhHQAAAAAAF0FIBwAAAADARRDSAQAAAABwEYR0AAAAAABcBCEdAAAAAAAXQUgHAAAAAMBFENIBAAAAAHARhHQAAAAAAFwEIR0AAAAAABfhEiE9MTFRkZGR8vHxUevWrbVly5Zrrv/JJ5+oXr168vHxUaNGjbRq1aoSqhQAAAAAgOLj9JC+ZMkSxcfHa9y4cdqxY4eaNGmi6OhonT59Ot/1N27cqH79+umhhx7STz/9pJ49e6pnz57as2dPCVcOAAAAAIBjOT2kT58+XUOHDlVsbKwaNGig2bNny8/PT3Pnzs13/TfffFNdu3bVqFGjVL9+fb388stq3ry5Zs6cWcKVAwAAAADgWB7O3HlmZqa2b9+uMWPGWNvc3NzUuXNnbdq0Kd9tNm3apPj4eJu26Ohoffrpp/mun5GRoYyMDOv71NRUSdLZs2eVlZVVxCMoXufPn1dmZqYsyUdlupha6H4sf51UTmamfj+bovPm9EL389dlszIzM3X+/Hl5eXkVuh9XwhiXDMa5ZDDOJYNxLhmMc8lgnEsG41wyGOeSwTgXzoULFyRJhmFcf2XDiU6cOGFIMjZu3GjTPmrUKKNVq1b5buPp6WksWrTIpi0xMdEIDQ3Nd/1x48YZknjx4sWLFy9evHjx4sWLFy+nvo4fP37dnOzUM+klYcyYMTZn3i0Wi86ePaubbrpJJpPJiZWVnLS0NFWrVk3Hjx9XYGCgs8splRjjksE4lwzGuWQwziWDcS4ZjHPJYJxLBuNcMsraOBuGoQsXLqhy5crXXdepIb1ixYpyd3dXcnKyTXtycrLCw8Pz3SY8PNyu9b29veXt7W3TFhwcXPiib2CBgYFl4j8AZ2KMSwbjXDIY55LBOJcMxrlkMM4lg3EuGYxzyShL4xwUFFSg9Zz64DgvLy+1aNFCSUlJ1jaLxaKkpCS1adMm323atGljs74kffPNN1ddHwAAAACAG4XTL3ePj49XTEyMWrZsqVatWikhIUHp6emKjY2VJA0aNEhVqlTR5MmTJUlPPfWUOnbsqGnTpql79+5avHixtm3bpnfffdeZhwEAAAAAQJE5PaT37dtXKSkpGjt2rE6dOqWmTZtq9erVCgsLkyQdO3ZMbm7/d8K/bdu2WrRokV544QU999xzql27tj799FPdcsstzjoEl+ft7a1x48bluewfjsMYlwzGuWQwziWDcS4ZjHPJYJxLBuNcMhjnksE4X53JMAryDHgAAAAAAFDcnHpPOgAAAAAA+D+EdAAAAAAAXAQhHQAAAAAAF0FIBwAAAADARRDSS7nExERFRkbKx8dHrVu31pYtW5xdUqnz7bffqkePHqpcubJMJpM+/fRTZ5dU6kyePFm33nqrAgICFBoaqp49e2r//v3OLqvUmTVrlho3bqzAwEAFBgaqTZs2+vLLL51dVqk3ZcoUmUwmjRgxwtmllCrjx4+XyWSyedWrV8/ZZZVKJ06c0IMPPqibbrpJvr6+atSokbZt2+bsskqVyMjIPN9nk8mk4cOHO7u0UiUnJ0cvvviibr75Zvn6+qpmzZp6+eWXxXO2HevChQsaMWKEIiIi5Ovrq7Zt22rr1q3OLsulENJLsSVLlig+Pl7jxo3Tjh071KRJE0VHR+v06dPOLq1USU9PV5MmTZSYmOjsUkqtDRs2aPjw4dq8ebO++eYbZWVlqUuXLkpPT3d2aaVK1apVNWXKFG3fvl3btm3THXfcoXvvvVe//PKLs0srtbZu3ap33nlHjRs3dnYppVLDhg31559/Wl/ff/+9s0sqdc6dO6d27drJ09NTX375pX799VdNmzZN5cuXd3ZppcrWrVttvsvffPONJOm+++5zcmWly9SpUzVr1izNnDlTe/fu1dSpU/Xqq69qxowZzi6tVHn44Yf1zTffaMGCBfr555/VpUsXde7cWSdOnHB2aS6DKdhKsdatW+vWW2/VzJkzJUkWi0XVqlXTE088odGjRzu5utLJZDJpxYoV6tmzp7NLKdVSUlIUGhqqDRs2qEOHDs4up1SrUKGCXnvtNT300EPOLqXUuXjxopo3b663335bEydOVNOmTZWQkODsskqN8ePH69NPP9XOnTudXUqpNnr0aP3www/67rvvnF1KmTJixAitXLlSv/32m0wmk7PLKTXuvvtuhYWF6b///a+1rXfv3vL19dWHH37oxMpKj0uXLikgIECfffaZunfvbm1v0aKF7rrrLk2cONGJ1bkOzqSXUpmZmdq+fbs6d+5sbXNzc1Pnzp21adMmJ1YGFF1qaqqkKwESxSMnJ0eLFy9Wenq62rRp4+xySqXhw4ere/fuNr+n4Vi//fabKleurBo1amjAgAE6duyYs0sqdf73v/+pZcuWuu+++xQaGqpmzZppzpw5zi6rVMvMzNSHH36oIUOGENAdrG3btkpKStKBAwckSbt27dL333+vu+66y8mVlR7Z2dnKycmRj4+PTbuvry9XO/2Nh7MLQPE4c+aMcnJyFBYWZtMeFhamffv2OakqoOgsFotGjBihdu3a6ZZbbnF2OaXOzz//rDZt2ujy5csqV66cVqxYoQYNGji7rFJn8eLF2rFjB/fgFaPWrVtr/vz5qlu3rv78809NmDBB7du31549exQQEODs8kqNQ4cOadasWYqPj9dzzz2nrVu36sknn5SXl5diYmKcXV6p9Omnn+r8+fMaPHiws0spdUaPHq20tDTVq1dP7u7uysnJ0SuvvKIBAwY4u7RSIyAgQG3atNHLL7+s+vXrKywsTB999JE2bdqkWrVqObs8l0FIB3BDGT58uPbs2cNfW4tJ3bp1tXPnTqWmpmrp0qWKiYnRhg0bCOoOdPz4cT311FP65ptv8pxJgOP8/cxX48aN1bp1a0VEROjjjz/m9g0HslgsatmypSZNmiRJatasmfbs2aPZs2cT0ovJf//7X911112qXLmys0spdT7++GMtXLhQixYtUsOGDbVz506NGDFClStX5vvsQAsWLNCQIUNUpUoVubu7q3nz5urXr5+2b9/u7NJcBiG9lKpYsaLc3d2VnJxs056cnKzw8HAnVQUUTVxcnFauXKlvv/1WVatWdXY5pZKXl5f1L9ktWrTQ1q1b9eabb+qdd95xcmWlx/bt23X69Gk1b97c2paTk6Nvv/1WM2fOVEZGhtzd3Z1YYekUHBysOnXq6ODBg84upVSpVKlSnj/i1a9fX8uWLXNSRaXb0aNHtWbNGi1fvtzZpZRKo0aN0ujRo/XAAw9Ikho1aqSjR49q8uTJhHQHqlmzpjZs2KD09HSlpaWpUqVK6tu3r2rUqOHs0lwG96SXUl5eXmrRooWSkpKsbRaLRUlJSdxfihuOYRiKi4vTihUrtHbtWt18883OLqnMsFgsysjIcHYZpUqnTp30888/a+fOndZXy5YtNWDAAO3cuZOAXkwuXryo33//XZUqVXJ2KaVKu3bt8kyJeeDAAUVERDipotJt3rx5Cg0NtXngFhzHbDbLzc02Hrm7u8tisTipotLN399flSpV0rlz5/TVV1/p3nvvdXZJLoMz6aVYfHy8YmJi1LJlS7Vq1UoJCQlKT09XbGyss0srVS5evGhzZubw4cPauXOnKlSooOrVqzuxstJj+PDhWrRokT777DMFBATo1KlTkqSgoCD5+vo6ubrSY8yYMbrrrrtUvXp1XbhwQYsWLdL69ev11VdfObu0UiUgICDP8xT8/f1100038ZwFBxo5cqR69OihiIgInTx5UuPGjZO7u7v69evn7NJKlf/85z9q27atJk2apPvvv19btmzRu+++q3fffdfZpZU6FotF8+bNU0xMjDw8+Cd8cejRo4deeeUVVa9eXQ0bNtRPP/2k6dOna8iQIc4urVT56quvZBiG6tatq4MHD2rUqFGqV68eGeXvDJRqM2bMMKpXr254eXkZrVq1MjZv3uzskkqddevWGZLyvGJiYpxdWqmR3/hKMubNm+fs0kqVIUOGGBEREYaXl5cREhJidOrUyfj666+dXVaZ0LFjR+Opp55ydhmlSt++fY1KlSoZXl5eRpUqVYy+ffsaBw8edHZZpdLnn39u3HLLLYa3t7dRr149491333V2SaXSV199ZUgy9u/f7+xSSq20tDTjqaeeMqpXr274+PgYNWrUMJ5//nkjIyPD2aWVKkuWLDFq1KhheHl5GeHh4cbw4cON8+fPO7ssl8I86QAAAAAAuAjuSQcAAAAAwEUQ0gEAAAAAcBGEdAAAAAAAXAQhHQAAAAAAF0FIBwAAAADARRDSAQAAAABwEYR0AAAAAABcBCEdAAAAAAAXQUgHAADFJioqSiaTSYMHD3Z2KQAA3BAI6QAA3EDOnTunMWPGqEGDBvL19VW5cuXUtGlTTZw4UWaz2Sk1HTlyRCaTSSaTSevXr7dZ1qBBA7Vu3Vo1a9Z0Sm0AANxoTIZhGM4uAgAAXN/JkyfVrl07HTlyRJIUGRmpzMxMnTx5UpLUtGlTbdiwQYGBgSVa15EjR3TzzTdLktatW6eoqKgS3T8AAKUJZ9IBALhBDBs2zBrQP/roIx0+fFgnTpzQ5MmTJUk7d+7U888/L0nWM9vz58+3bp/fpeepqal66qmnFBERIS8vL1WtWlXx8fE2Z+X379+ve+65R6GhofL29lbVqlV11113acuWLZo/f741oEvS7bffLpPJZA3q+e3z7NmzGj58uKpVqyZPT0+FhYXpwQcf1LFjx6zrjB8/XiaTSZGRkfrkk09Ur149+fv7q0OHDtq/f7+DRhQAANdDSAcA4AZw/vx5ff7555KuBN8HHnjAuuyZZ56xBuVFixapoBfJZWZmKioqSm+99ZZOnz6t+vXr66+//tIbb7yhHj16WPvp16+fPv/8c2VnZ6thw4ayWCxavXq1fv31V4WEhKhp06bWPuvXr6/WrVurQYMG+e7z8uXL6tixo95++22dOnVKderUUVpamhYuXKg2bdooJSXFZv0TJ05owIABMplMunTpkr777jsNGTKkwOMGAMCNhpAOAMAN4MCBA7JYLJJkE4olyc3NTY0bN5Z05Sz1mTNnCtTnRx99pJ07d8rLy0u7d+/Wrl27tHnzZknS2rVrtXbtWknSb7/9Jkn6/PPPtWPHDp08eVKHDh1SVFSUunfvrhUrVlj7fPvtt7V582a9/fbbV93nnj17JEmffPKJfvnlF/3www9yc3PTyZMnNXPmTJv1s7OztWzZMu3du1cjRoyQJG3cuFGXLl0q0DECAHCjIaQDAHCDcXPL+7/vv7d5enoWqJ8tW7ZIunJGvU6dOjKZTDZ/AMgN7D169JB05VL2+vXrq3fv3lq9erUqVapkd+1bt26VJPn5+alnz56SpObNm6tu3bqSpG3bttmsHxQUZN3/38/Onz592u59AwBwI/BwdgEAAOD6atWqJZPJJMMwtHv3bptlFotFu3btkiSFh4crODjYuiwnJ8f6c2pqar59e3l5qVmzZnnay5cvL0n64IMPdM8992j9+vX69ddftWrVKi1fvlx79uxRYmJiUQ/tmv5+LB4e//fPFp57CwAorTiTDgDADaBChQrq1q2bJCkpKUmfffaZddmrr76qQ4cOSZKGDh0qSQoNDZV05TJ5Sdq3b59+/vlnmz5vvfVWSVeCfO5l6ps3b9b69es1atQo9e/fX5L03XffqVevXpo9e7a+/fZbjRs3TpL07bffSrpyVjxXenr6NY8jd59ms1mffvqpJGnHjh3Wh8G1bNmywGMCAEBpREgHAOAG8fbbb6tq1aoyDEM9e/ZUzZo1VaVKFY0ZM0aS1LFjR+vT3Tt16iRJmjZtmm6//Xa1adMmz9nnfv36qXHjxsrJydGtt96qW265RXXr1lVwcLD69Omj8+fPS5IGDhyo8uXLq27dumrWrJnGjh0rSdb74ENCQnTTTTdZ123durVmzJiR7zH069dPt9xyiyTpvvvuU8OGDdWuXTtZLBZVrlxZcXFxDhwxAABuPIR0AABuENWrV9dPP/2kZ555RvXr19fJkyetc6Q/+uijWrNmjby9vSVJ06dPV/fu3eXr66vff/9dzz33nG677Tab/ry9vbVhwwY9+eSTqlatmg4cOKBz586pZcuWeuWVVxQWFiZJio2NVcOGDXXmzBn9+uuvCg8P1yOPPGJ9yJvJZNKcOXNUq1YtpaWlacuWLTp69Gi+x+Dj46MNGzZo2LBhCg8P14EDBxQQEKABAwZo06ZNCgkJKa7hAwDghmAyuKkLAIAb1pw5c/TII48oPDxcmzdvVkREhLNLAgAARcCZdAAAbmBDhw7VY489plOnTql79+5XfTgcAAC4MXAmHQAAAAAAF8GZdAAAAAAAXAQhHQAAAAAAF0FIBwAAAADARRDSAQAAAABwEYR0AAAAAABcBCEdAAAAAAAXQUgHAAAAAMBFENIBAAAAAHARhHQAAAAAAFzE/wOOSBpFTkc0MAAAAABJRU5ErkJggg==",
      "text/plain": [
       "<Figure size 1200x600 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "question_numbers = np.arange(len(data))  # Question numbers for X-axis\n",
    "\n",
    "plt.figure(figsize=(12, 6))\n",
    "bar_width = 0.2  # Width of the bars\n",
    "\n",
    "# Set position of bar on X axis\n",
    "r1 = np.arange(len(simulated_scores_gpt4))\n",
    "r2 = [x + bar_width for x in r1]\n",
    "\n",
    "# Make the plot with pastel colors\n",
    "plt.bar(r1, ldcc_similarity_scores, color='skyblue', width=bar_width, edgecolor='grey', label='ldcc')\n",
    "plt.bar(r2, llama_similarity_scores, color='lightpink', width=bar_width, edgecolor='grey', label='llama')\n",
    "\n",
    "# Adjust xticks to be in the middle of the group bars\n",
    "plt.xlabel('Question', fontweight='bold')\n",
    "plt.xticks([r + bar_width / 2 for r in range(len(simulated_scores_gpt4))], question_numbers)\n",
    "plt.ylabel('Cosine Similarity wrt Answer')\n",
    "plt.title('simeple_Q Comparing local models, TASK : RAG')\n",
    "plt.grid(axis='y') \n",
    "# Create legend & Show graphic\n",
    "plt.legend()\n",
    "plt.show()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 74,
   "id": "97f27107-b670-46bb-a418-4a926baa691c",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1400x700 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "average_score_ldcc_similarity_scores = np.mean(ldcc_similarity_scores)\n",
    "average_score_llama_similarity_scores = np.mean(llama_similarity_scores)\n",
    "\n",
    "# Extend the question numbers array to include a slot for the average\n",
    "extended_question_numbers = np.append(question_numbers, [len(question_numbers)])  # Add an index for the average\n",
    "\n",
    "plt.figure(figsize=(14, 7))\n",
    "bar_width = 0.2  # Width of the bars\n",
    "\n",
    "# Set position of bar on X axis for the extended array\n",
    "r1_extended = np.arange(len(extended_question_numbers))\n",
    "r2_extended = [x + bar_width for x in r1_extended]\n",
    "\n",
    "# Plot the original scores\n",
    "plt.bar(r1, ldcc_similarity_scores, color='skyblue', width=bar_width, edgecolor='grey', label='ldcc')\n",
    "plt.bar(r2, llama_similarity_scores, color='lightpink', width=bar_width, edgecolor='grey', label='llama')\n",
    "\n",
    "# Plot the average scores as the last pair of bars\n",
    "plt.bar(len(extended_question_numbers)-1 - 0.2, average_score_ldcc_similarity_scores, color='darkblue', width=bar_width, edgecolor='grey')\n",
    "plt.bar(len(extended_question_numbers)-1, average_score_llama_similarity_scores, color='deeppink', width=bar_width, edgecolor='grey')\n",
    "\n",
    "# Adjust xticks to include the average\n",
    "plt.xlabel('Question', fontweight='bold')\n",
    "plt.xticks(r1_extended + bar_width / 2, list(question_numbers) + ['Average'])\n",
    "plt.ylabel('Simulated Cosine Similarity')\n",
    "plt.title('simeple_Q Comparing local models, TASK : RAG (Including Average)')\n",
    "plt.grid(axis='y')\n",
    "\n",
    "# Create legend & Show graphic\n",
    "plt.legend()\n",
    "plt.show()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "4bb415ea-cb70-4538-a3b6-bb41f22b950f",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "llm",
   "language": "python",
   "name": "llm"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.10.13"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}