02_Where_finetuning_fits_in.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "b944c304",
   "metadata": {},
   "source": [
    "# Finetuning data: compare to pretraining and basic preparation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "fec5e3b3",
   "metadata": {},
   "outputs": [],
   "source": [
    "import jsonlines\n",
    "import itertools\n",
    "import pandas as pd\n",
    "from pprint import pprint\n",
    "\n",
    "import datasets\n",
    "from datasets import load_dataset"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3cd7700c",
   "metadata": {},
   "source": [
    "## Look at pretraining data set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "bc92a955",
   "metadata": {},
   "outputs": [],
   "source": [
    "#pretrained_dataset = load_dataset(\"EleutherAI/pile\", split=\"train\", streaming=True)\n",
    "\n",
    "pretrained_dataset = load_dataset(\"c4\", \"en\", split=\"train\", streaming=True)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "5ae5f773",
   "metadata": {},
   "outputs": [],
   "source": [
    "n = 5\n",
    "print(\"Pretrained dataset:\")\n",
    "top_n = itertools.islice(pretrained_dataset, n)\n",
    "for i in top_n:\n",
    "  print(i)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "cc33338d",
   "metadata": {},
   "source": [
    "## Contrast with company finetuning dataset you will be using"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "66668b38",
   "metadata": {},
   "outputs": [],
   "source": [
    "filename = \"lamini_docs.jsonl\"\n",
    "instruction_dataset_df = pd.read_json(filename, lines=True)\n",
    "instruction_dataset_df"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b84160b8",
   "metadata": {},
   "source": [
    "## Various ways of formatting your data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "dca6b7e7",
   "metadata": {},
   "outputs": [],
   "source": [
    "examples = instruction_dataset_df.to_dict()\n",
    "text = examples[\"question\"][0] + examples[\"answer\"][0]\n",
    "text"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "5b0fae24",
   "metadata": {},
   "outputs": [],
   "source": [
    "if \"question\" in examples and \"answer\" in examples:\n",
    "  text = examples[\"question\"][0] + examples[\"answer\"][0]\n",
    "elif \"instruction\" in examples and \"response\" in examples:\n",
    "  text = examples[\"instruction\"][0] + examples[\"response\"][0]\n",
    "elif \"input\" in examples and \"output\" in examples:\n",
    "  text = examples[\"input\"][0] + examples[\"output\"][0]\n",
    "else:\n",
    "  text = examples[\"text\"][0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "cc40e0ab",
   "metadata": {},
   "outputs": [],
   "source": [
    "prompt_template_qa = \"\"\"### Question:\n",
    "{question}\n",
    "\n",
    "### Answer:\n",
    "{answer}\"\"\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "055b5661",
   "metadata": {},
   "outputs": [],
   "source": [
    "question = examples[\"question\"][0]\n",
    "answer = examples[\"answer\"][0]\n",
    "\n",
    "text_with_prompt_template = prompt_template_qa.format(question=question, answer=answer)\n",
    "text_with_prompt_template"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "11b1923b",
   "metadata": {},
   "outputs": [],
   "source": [
    "prompt_template_q = \"\"\"### Question:\n",
    "{question}\n",
    "\n",
    "### Answer:\"\"\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "f81a6078",
   "metadata": {},
   "outputs": [],
   "source": [
    "num_examples = len(examples[\"question\"])\n",
    "finetuning_dataset_text_only = []\n",
    "finetuning_dataset_question_answer = []\n",
    "for i in range(num_examples):\n",
    "  question = examples[\"question\"][i]\n",
    "  answer = examples[\"answer\"][i]\n",
    "\n",
    "  text_with_prompt_template_qa = prompt_template_qa.format(question=question, answer=answer)\n",
    "  finetuning_dataset_text_only.append({\"text\": text_with_prompt_template_qa})\n",
    "\n",
    "  text_with_prompt_template_q = prompt_template_q.format(question=question)\n",
    "  finetuning_dataset_question_answer.append({\"question\": text_with_prompt_template_q, \"answer\": answer})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "f517adf4",
   "metadata": {},
   "outputs": [],
   "source": [
    "pprint(finetuning_dataset_text_only[0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "5febb8c8",
   "metadata": {},
   "outputs": [],
   "source": [
    "pprint(finetuning_dataset_question_answer[0])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7e93258d",
   "metadata": {},
   "source": [
    "## Common ways of storing your data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "38e91b0a",
   "metadata": {},
   "outputs": [],
   "source": [
    "with jsonlines.open(f'lamini_docs_processed.jsonl', 'w') as writer:\n",
    "    writer.write_all(finetuning_dataset_question_answer)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "517cfc56",
   "metadata": {},
   "outputs": [],
   "source": [
    "finetuning_dataset_name = \"lamini/lamini_docs\"\n",
    "finetuning_dataset = load_dataset(finetuning_dataset_name)\n",
    "print(finetuning_dataset)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.16"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}