{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from datasets import load_dataset, load_from_disk"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "boolq = load_dataset(\"google/boolq\")\n",
    "boolq"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "commonsense_boolq = []\n",
    "for data in boolq['train']:\n",
    "    question = data['question']\n",
    "    answer = {True:'true', False:'false'}[data['answer']]\n",
    "    commonsense_boolq.append(\n",
    "        {\n",
    "            'instruction': f\"Please answer the following question with true or false, question: {question}?\\n\\nAnswer format: true/false\",\n",
    "            'answer': answer,\n",
    "            'input': '',\n",
    "            'output': f'the correct answer is {answer}'\n",
    "        }\n",
    "    )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "piqa = load_dataset(\"skrishna/piqa_preop\")\n",
    "piqa"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "commonsense_piqa = []\n",
    "for data in piqa['train']:\n",
    "    goal = data['goal']\n",
    "    sol1 = data['sol1']\n",
    "    sol2 = data['sol2']\n",
    "    label = data['label']+1\n",
    "    commonsense_piqa.append(\n",
    "        {\n",
    "            'instruction': f\"Please choose the correct solution to the question: {goal}\\n\\nSolution1: {sol1}\\n\\nSolution2: {sol2}\\n\\nAnswer format: solution1/solution2\",\n",
    "            'answer': f\"solution{label}\",\n",
    "            'input': '',\n",
    "            'output': f'the correct answer is solution{label}'\n",
    "        }\n",
    "    )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "siqa = load_dataset(\"lighteval/siqa\")\n",
    "siqa"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "commonsense_siqa = []\n",
    "for data in siqa['train']:\n",
    "    context = data['context']\n",
    "    question = data['question']\n",
    "    answerA = data['answerA']\n",
    "    answerB = data['answerB']\n",
    "    answerC = data['answerC']\n",
    "    label = data['label']\n",
    "    commonsense_siqa.append(\n",
    "        {\n",
    "            'instruction': f\"Please choose the correct answer to the question: {context} {question}\\n\\nAnswer1: {answerA} Answer2: {answerB} Answer3: {answerC}\\n\\nAnswer format: answer1/answer2/answer3\",\n",
    "            'answer': f\"answer{label}\",\n",
    "            'input': '',\n",
    "            'output': f'the correct answer is answer{label}'\n",
    "        }\n",
    "    )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "hellaswag = load_dataset(\"/Users/mengfanxu/hellaswag/hellaswag_train\")\n",
    "hellaswag"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "commonsense_hellaswag = []\n",
    "for data in hellaswag['train']:\n",
    "    activity_label = data['activity_label']\n",
    "    ctx = data['ctx']\n",
    "    ed1 = data['endings'][0]\n",
    "    ed2 = data['endings'][1]\n",
    "    ed3 = data['endings'][2]\n",
    "    ed4 = data['endings'][3]\n",
    "    label = data['label']+1\n",
    "    commonsense_hellaswag.append(\n",
    "        {\n",
    "            'instruction': f\"Please choose the correct ending to complete the given sentence: {activity_label}: {ctx}\\n\\nEnding1: {ed1} Ending2: {ed2} Ending3: {ed3} Ending4: {ed4}\\n\\nAnswer format: ending1/ending2/ending3/ending4\",\n",
    "            'answer': f\"ending{label}\",\n",
    "            'input': '',\n",
    "            'output': f'the correct answer is ending{label}'\n",
    "        }\n",
    "    )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "openbookqa = load_from_disk(\"/Users/mengfanxu/openbookqa\")\n",
    "openbookqa"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "commonsense_openbookqa = []\n",
    "for data in openbookqa['train']:\n",
    "    question_stem = data['question_stem']\n",
    "    ed1 = data['choices']['text'][0]\n",
    "    ed2 = data['choices']['text'][1]\n",
    "    ed3 = data['choices']['text'][2]\n",
    "    ed4 = data['choices']['text'][3]\n",
    "    label = {\"A\":1,\"B\":2,\"C\":3,\"D\":4}[data['answerKey']]\n",
    "    commonsense_openbookqa.append(\n",
    "        {\n",
    "            'instruction': f\"Please choose the correct answer to the question: {question_stem}\\n\\nAnswer1: {ed1} Answer2: {ed2} Answer3: {ed3} Answer4: {ed4}\\n\\nAnswer format: answer1/answer2/answer3/answer4\",\n",
    "            'answer': f\"answer{label}\",\n",
    "            'input': '',\n",
    "            'output': f'the correct answer is answer{label}'\n",
    "        }\n",
    "    )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "arc_c = load_dataset(\"allenai/ai2_arc\", \"ARC-Challenge\")\n",
    "arc_c"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "commonsense_arc_c = []\n",
    "for data in arc_c['train']:\n",
    "    question = data['question']\n",
    "    ed1 = data['choices']['text'][0]\n",
    "    ed2 = data['choices']['text'][1]\n",
    "    ed3 = data['choices']['text'][2]\n",
    "    instruction = f\"Please choose the correct answer to the question: {question}\\n\\nAnswer1: {ed1} Answer2: {ed2} Answer3: {ed3}\\n\\nAnswer format: answer1/answer2/answer3\"\n",
    "    if len(data['choices']['text'])>=4:\n",
    "        ed4 = data['choices']['text'][3]\n",
    "        instruction = f\"Please choose the correct answer to the question: {question}\\n\\nAnswer1: {ed1} Answer2: {ed2} Answer3: {ed3} Answer4: {ed4}\\n\\nAnswer format: answer1/answer2/answer3/answer4\"\n",
    "    if len(data['choices']['text'])>=5:\n",
    "        ed5 = data['choices']['text'][4]\n",
    "        instruction = f\"Please choose the correct answer to the question: {question}\\n\\nAnswer1: {ed1} Answer2: {ed2} Answer3: {ed3} Answer4: {ed4} Answer5: {ed5}\\n\\nAnswer format: answer1/answer2/answer3/answer4/answer5\"\n",
    "    label = {\"A\":1,\"B\":2,\"C\":3,\"D\":4,\"E\":5, \"2\":2, \"4\":4, \"1\":1, \"3\":3}[data['answerKey']]\n",
    "    \n",
    "    commonsense_arc_c.append(\n",
    "        {\n",
    "            'instruction': instruction,\n",
    "            'answer': f\"answer{label}\",\n",
    "            'input': '',\n",
    "            'output': f'the correct answer is answer{label}'\n",
    "        }\n",
    "    )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "arc_e = load_dataset(\"allenai/ai2_arc\", \"ARC-Easy\")\n",
    "arc_e"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "commonsense_arc_e = []\n",
    "for data in arc_e['train']:\n",
    "    question = data['question']\n",
    "    ed1 = data['choices']['text'][0]\n",
    "    ed2 = data['choices']['text'][1]\n",
    "    ed3 = data['choices']['text'][2]\n",
    "    instruction = f\"Please choose the correct answer to the question: {question}\\n\\nAnswer1: {ed1} Answer2: {ed2} Answer3: {ed3}\\n\\nAnswer format: answer1/answer2/answer3\"\n",
    "    if len(data['choices']['text'])>=4:\n",
    "        ed4 = data['choices']['text'][3]\n",
    "        instruction = f\"Please choose the correct answer to the question: {question}\\n\\nAnswer1: {ed1} Answer2: {ed2} Answer3: {ed3} Answer4: {ed4}\\n\\nAnswer format: answer1/answer2/answer3/answer4\"\n",
    "    if len(data['choices']['text'])>=5:\n",
    "        ed5 = data['choices']['text'][4]\n",
    "        instruction = f\"Please choose the correct answer to the question: {question}\\n\\nAnswer1: {ed1} Answer2: {ed2} Answer3: {ed3} Answer4: {ed4} Answer5: {ed5}\\n\\nAnswer format: answer1/answer2/answer3/answer4/answer5\"\n",
    "    label = {\"A\":1,\"B\":2,\"C\":3,\"D\":4,\"E\":5, \"2\":2, \"4\":4, \"1\":1, \"3\":3}[data['answerKey']]\n",
    "    \n",
    "    commonsense_arc_e.append(\n",
    "        {\n",
    "            'instruction': instruction,\n",
    "            'answer': f\"answer{label}\",\n",
    "            'input': '',\n",
    "            'output': f'the correct answer is answer{label}'\n",
    "        }\n",
    "    )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "winogrande = load_dataset(\"/Users/mengfanxu/Downloads/winogrande_1.1/train\")\n",
    "winogrande"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "commonsense_winogrande = []\n",
    "for data in winogrande['train']:\n",
    "    sentence = data['sentence']\n",
    "    option1 = data['option1']\n",
    "    option2 = data['option2']\n",
    "    answer = data['answer']\n",
    "\n",
    "    commonsense_winogrande.append(\n",
    "        {\n",
    "            'instruction': f\"Please choose the correct answer to fill in the blank to complete the given sentence: {sentence}\\n\\nOption1: {option1} Option2: {option2} Answer format: option1/option2\",\n",
    "            'answer': f\"option{answer}\",\n",
    "            'input': '',\n",
    "            'output': f'the correct answer is option{answer}'\n",
    "        }\n",
    "    )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "eval_boolq = load_from_disk(\"/Users/mengfanxu/Downloads/winogrande_1.1/PiSSA/inference/data/eval_boolq\")\n",
    "eval_piqa = load_from_disk(\"/Users/mengfanxu/Downloads/winogrande_1.1/PiSSA/inference/data/eval_piqa\")\n",
    "eval_social_interaction_qa = load_from_disk(\"/Users/mengfanxu/Downloads/winogrande_1.1/PiSSA/inference/data/eval_social_interaction_qa\")\n",
    "eval_hellaswag = load_from_disk(\"/Users/mengfanxu/Downloads/winogrande_1.1/PiSSA/inference/data/eval_hellaswag\")\n",
    "eval_winogrande = load_from_disk(\"/Users/mengfanxu/Downloads/winogrande_1.1/PiSSA/inference/data/eval_winogrande\")\n",
    "eval_arc_challenge = load_from_disk(\"/Users/mengfanxu/Downloads/winogrande_1.1/PiSSA/inference/data/eval_arc_challenge\")\n",
    "eval_arc_easy = load_from_disk(\"/Users/mengfanxu/Downloads/winogrande_1.1/PiSSA/inference/data/eval_arc_easy\")\n",
    "eval_openbookqa = load_from_disk(\"/Users/mengfanxu/Downloads/winogrande_1.1/PiSSA/inference/data/eval_openbookqa\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import json\n",
    "\n",
    "with open(\"commonsense_filtered/boolq/train.json\", \"w\", encoding=\"utf-8\") as f:\n",
    "    for item in commonsense_boolq:\n",
    "        f.write(json.dumps(item, ensure_ascii=False) + \"\\n\")\n",
    "\n",
    "with open(\"commonsense_filtered/boolq/test.json\", \"w\", encoding=\"utf-8\") as f:\n",
    "    for item in eval_boolq:\n",
    "        f.write(json.dumps(item, ensure_ascii=False) + \"\\n\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import json\n",
    "\n",
    "with open(\"commonsense_filtered/piqa/train.json\", \"w\", encoding=\"utf-8\") as f:\n",
    "    for item in commonsense_piqa:\n",
    "        f.write(json.dumps(item, ensure_ascii=False) + \"\\n\")\n",
    "\n",
    "with open(\"commonsense_filtered/piqa/test.json\", \"w\", encoding=\"utf-8\") as f:\n",
    "    for item in eval_piqa:\n",
    "        f.write(json.dumps(item, ensure_ascii=False) + \"\\n\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import json\n",
    "\n",
    "with open(\"commonsense_filtered/siqa/train.json\", \"w\", encoding=\"utf-8\") as f:\n",
    "    for item in commonsense_siqa:\n",
    "        f.write(json.dumps(item, ensure_ascii=False) + \"\\n\")\n",
    "\n",
    "with open(\"commonsense_filtered/siqa/test.json\", \"w\", encoding=\"utf-8\") as f:\n",
    "    for item in eval_social_interaction_qa:\n",
    "        f.write(json.dumps(item, ensure_ascii=False) + \"\\n\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import json\n",
    "\n",
    "with open(\"commonsense_filtered/hellaswag/train.json\", \"w\", encoding=\"utf-8\") as f:\n",
    "    for item in commonsense_hellaswag:\n",
    "        f.write(json.dumps(item, ensure_ascii=False) + \"\\n\")\n",
    "\n",
    "with open(\"commonsense_filtered/hellaswag/test.json\", \"w\", encoding=\"utf-8\") as f:\n",
    "    for item in eval_hellaswag:\n",
    "        f.write(json.dumps(item, ensure_ascii=False) + \"\\n\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import json\n",
    "\n",
    "with open(\"commonsense_filtered/winogrande/train.json\", \"w\", encoding=\"utf-8\") as f:\n",
    "    for item in commonsense_winogrande:\n",
    "        f.write(json.dumps(item, ensure_ascii=False) + \"\\n\")\n",
    "\n",
    "with open(\"commonsense_filtered/winogrande/test.json\", \"w\", encoding=\"utf-8\") as f:\n",
    "    for item in eval_winogrande:\n",
    "        f.write(json.dumps(item, ensure_ascii=False) + \"\\n\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import json\n",
    "\n",
    "with open(\"commonsense_filtered/arc_challenge/train.json\", \"w\", encoding=\"utf-8\") as f:\n",
    "    for item in commonsense_arc_c:\n",
    "        f.write(json.dumps(item, ensure_ascii=False) + \"\\n\")\n",
    "\n",
    "with open(\"commonsense_filtered/arc_challenge/test.json\", \"w\", encoding=\"utf-8\") as f:\n",
    "    for item in eval_arc_challenge:\n",
    "        f.write(json.dumps(item, ensure_ascii=False) + \"\\n\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import json\n",
    "\n",
    "with open(\"commonsense_filtered/arc_easy/train.json\", \"w\", encoding=\"utf-8\") as f:\n",
    "    for item in commonsense_arc_e:\n",
    "        f.write(json.dumps(item, ensure_ascii=False) + \"\\n\")\n",
    "\n",
    "with open(\"commonsense_filtered/arc_easy/test.json\", \"w\", encoding=\"utf-8\") as f:\n",
    "    for item in eval_arc_easy:\n",
    "        f.write(json.dumps(item, ensure_ascii=False) + \"\\n\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import json\n",
    "\n",
    "with open(\"commonsense_filtered/openbookqa/train.json\", \"w\", encoding=\"utf-8\") as f:\n",
    "    for item in commonsense_openbookqa:\n",
    "        f.write(json.dumps(item, ensure_ascii=False) + \"\\n\")\n",
    "\n",
    "with open(\"commonsense_filtered/openbookqa/test.json\", \"w\", encoding=\"utf-8\") as f:\n",
    "    for item in eval_openbookqa:\n",
    "        f.write(json.dumps(item, ensure_ascii=False) + \"\\n\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "base",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.16"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}