all_prompt_engineering_techniques/constrained-guided-generation.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Constrained and Guided Generation Tutorial\n",
    "\n",
    "## Overview\n",
    "\n",
    "This tutorial explores the concepts of constrained and guided generation in the context of large language models. We'll focus on techniques to set up constraints for model outputs and implement rule-based generation using OpenAI's GPT models and the LangChain library.\n",
    "\n",
    "## Motivation\n",
    "\n",
    "While large language models are powerful tools for generating text, they sometimes produce outputs that are too open-ended or lack specific desired characteristics. Constrained and guided generation techniques allow us to exert more control over the model's outputs, making them more suitable for specific tasks or adhering to certain rules and formats.\n",
    "\n",
    "## Key Components\n",
    "\n",
    "1. Setting up constraints for model outputs\n",
    "2. Implementing rule-based generation\n",
    "3. Using LangChain's PromptTemplate for structured prompts\n",
    "4. Leveraging OpenAI's GPT models for text generation\n",
    "\n",
    "## Method Details\n",
    "\n",
    "We'll use a combination of prompt engineering techniques and LangChain's utilities to implement constrained and guided generation:\n",
    "\n",
    "1. We'll start by setting up the environment and importing necessary libraries.\n",
    "2. We'll create structured prompts using LangChain's PromptTemplate to guide the model's output.\n",
    "3. We'll implement constraints by specifying rules and formats in our prompts.\n",
    "4. We'll use OpenAI's GPT model to generate text based on our constrained prompts.\n",
    "5. We'll explore different techniques for rule-based generation, including output parsing and regex-based validation.\n",
    "\n",
    "## Conclusion\n",
    "\n",
    "By the end of this tutorial, you'll have a solid understanding of how to implement constrained and guided generation techniques. These skills will enable you to create more controlled and specific outputs from large language models, making them more suitable for a wide range of applications where precise and rule-adherent text generation is required."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Setup\n",
    "\n",
    "First, let's import the necessary libraries and set up our environment."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import re\n",
    "\n",
    "from langchain_openai import ChatOpenAI\n",
    "from langchain.prompts import PromptTemplate\n",
    "from langchain.output_parsers import RegexParser\n",
    "\n",
    "from dotenv import load_dotenv\n",
    "load_dotenv()\n",
    "\n",
    "# Set up the OpenAI API key\n",
    "os.environ[\"OPENAI_API_KEY\"] = os.getenv('OPENAI_API_KEY')\n",
    "\n",
    "# Initialize the language model\n",
    "llm = ChatOpenAI(model=\"gpt-4o-mini\")\n",
    "\n",
    "# Function to display model outputs\n",
    "def display_output(output):\n",
    "    \"\"\"Display the model's output in a formatted manner.\"\"\"\n",
    "    print(\"Model Output:\")\n",
    "    print(\"-\" * 40)\n",
    "    print(output)\n",
    "    print(\"-\" * 40)\n",
    "    print()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Setting Up Constraints for Model Outputs\n",
    "\n",
    "Let's start by creating a constrained prompt that generates a product description with specific requirements."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model Output:\n",
      "----------------------------------------\n",
      "**Stay Hydrated, Stay Awesome!**  \n",
      "\n",
      "Meet your new hydration buddy! Our Smart Water Bottle tracks your water intake, reminds you to sip throughout the day, and syncs with your favorite fitness apps. Made from eco-friendly materials and designed for on-the-go lifestyles, it’s the perfect accessory for health-conscious millennials. Ready to elevate your hydration game? Grab yours today and drink up the good vibes!\n",
      "----------------------------------------\n",
      "\n"
     ]
    }
   ],
   "source": [
    "constrained_prompt = PromptTemplate(\n",
    "    input_variables=[\"product\", \"target_audience\", \"tone\", \"word_limit\"],\n",
    "    template=\"\"\"Create a product description for {product} targeted at {target_audience}.\n",
    "    Use a {tone} tone and keep it under {word_limit} words.\n",
    "    The description should include:\n",
    "    1. A catchy headline\n",
    "    2. Three key features\n",
    "    3. A call to action\n",
    "    \n",
    "    Product Description:\n",
    "    \"\"\"\n",
    ")\n",
    "\n",
    "# Generate the constrained output\n",
    "input_variables = {\n",
    "    \"product\": \"smart water bottle\",\n",
    "    \"target_audience\": \"health-conscious millennials\",\n",
    "    \"tone\": \"casual and friendly\",\n",
    "    \"word_limit\": \"75\"\n",
    "}\n",
    "\n",
    "chain = constrained_prompt | llm\n",
    "output = chain.invoke(input_variables).content\n",
    "display_output(output)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Implementing Rule-Based Generation\n",
    "\n",
    "Now, let's implement a rule-based generation system for creating structured job postings."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model Output:\n",
      "----------------------------------------\n",
      "COMPANY: TechInnovate Solutions is a forward-thinking technology firm dedicated to developing cutting-edge software solutions that drive success for businesses worldwide. Located in the heart of San Francisco, we pride ourselves on fostering a collaborative and innovative work environment.\n",
      "\n",
      "RESPONSIBILITIES:\n",
      "- Design and implement robust software architectures to support scalable applications.\n",
      "- Lead cross-functional teams in the development and deployment of new features and enhancements.\n",
      "- Collaborate with product managers to define and prioritize product requirements.\n",
      "- Mentor junior engineers and provide guidance on best coding practices and methodologies.\n",
      "- Conduct code reviews and ensure adherence to industry standards and quality assurance processes.\n",
      "\n",
      "QUALIFICATIONS:\n",
      "- A minimum of 5 years of professional software engineering experience is required. \n",
      "- Proficiency in programming languages such as Java, Python, or JavaScript is essential.\n",
      "- Strong understanding of software development methodologies, including Agile and DevOps practices.\n",
      "- Experience with cloud platforms such as AWS, Azure, or Google Cloud is preferred.\n",
      "- Excellent problem-solving skills and the ability to work effectively in a team-oriented environment are necessary.\n",
      "\n",
      "EEO: TechInnovate Solutions is an equal opportunity employer. We celebrate diversity and are committed to creating an inclusive environment for all employees.\n",
      "----------------------------------------\n",
      "\n"
     ]
    }
   ],
   "source": [
    "job_posting_prompt = PromptTemplate(\n",
    "    input_variables=[\"job_title\", \"company\", \"location\", \"experience\"],\n",
    "    template=\"\"\"Create a job posting for a {job_title} position at {company} in {location}.\n",
    "    The candidate should have {experience} years of experience.\n",
    "    Follow these rules:\n",
    "    1. Start with a brief company description (2 sentences)\n",
    "    2. List 5 key responsibilities, each starting with an action verb\n",
    "    3. List 5 required qualifications, each in a single sentence\n",
    "    4. End with a standardized equal opportunity statement\n",
    "    \n",
    "    Format the output as follows:\n",
    "    COMPANY: [Company Description]\n",
    "    \n",
    "    RESPONSIBILITIES:\n",
    "    - [Responsibility 1]\n",
    "    - [Responsibility 2]\n",
    "    - [Responsibility 3]\n",
    "    - [Responsibility 4]\n",
    "    - [Responsibility 5]\n",
    "    \n",
    "    QUALIFICATIONS:\n",
    "    - [Qualification 1]\n",
    "    - [Qualification 2]\n",
    "    - [Qualification 3]\n",
    "    - [Qualification 4]\n",
    "    - [Qualification 5]\n",
    "    \n",
    "    EEO: [Equal Opportunity Statement]\n",
    "    \"\"\"\n",
    ")\n",
    "\n",
    "# Generate the rule-based output\n",
    "input_variables = {\n",
    "    \"job_title\": \"Senior Software Engineer\",\n",
    "    \"company\": \"TechInnovate Solutions\",\n",
    "    \"location\": \"San Francisco, CA\",\n",
    "    \"experience\": \"5+\"\n",
    "}\n",
    "\n",
    "chain = job_posting_prompt | llm\n",
    "output = chain.invoke(input_variables).content\n",
    "display_output(output)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Using Regex Parser for Structured Output\n",
    "\n",
    "Let's use a regex parser to ensure our output adheres to a specific structure."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Parsed Output:\n",
      "COMPANY_DESCRIPTION:\n",
      "TechInnovate Solutions is a leading technology firm based in San Francisco, CA, dedicated to creating cutting-edge software solutions that empower businesses to thrive in the digital age. Our team of innovative thinkers and problem solvers is committed to pushing the boundaries of technology to deliver exceptional products and services.\n",
      "\n",
      "RESPONSIBILITIES:\n",
      "- Design and develop scalable software applications that meet the needs of our clients.\n",
      "- Collaborate with cross-functional teams to define, design, and implement new features.\n",
      "- Mentor junior engineers, providing guidance and support for their professional growth.\n",
      "- Troubleshoot and resolve software defects and performance issues in a timely manner.\n",
      "- Stay updated with emerging technologies and industry trends to ensure best practices.\n",
      "\n",
      "QUALIFICATIONS:\n",
      "- A minimum of 5 years of experience in software development, with a strong focus on full-stack technologies. \n",
      "- Proficiency in programming languages such as Java, Python, or JavaScript, along with relevant frameworks.\n",
      "- Experience with cloud platforms such as AWS, Azure, or Google Cloud.\n",
      "- Strong understanding of software development life cycle (SDLC) and agile methodologies.\n",
      "- Excellent problem-solving skills and ability to work in a fast-paced environment.\n",
      "\n",
      "EEO_STATEMENT:\n",
      "TechInnovate Solutions is an equal opportunity employer. We celebrate diversity and are committed to creating an inclusive environment for all employees.\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# Define a regex parser for structured output\n",
    "regex_parser = RegexParser(\n",
    "    regex=r\"COMPANY:\\s*([\\s\\S]*?)\\n\\s*RESPONSIBILITIES:\\s*([\\s\\S]*?)\\n\\s*QUALIFICATIONS:\\s*([\\s\\S]*?)\\n\\s*EEO:\\s*([\\s\\S]*)\",\n",
    "    output_keys=[\"company_description\", \"responsibilities\", \"qualifications\", \"eeo_statement\"]\n",
    ")\n",
    "# This regex pattern captures the company description, responsibilities, qualifications, and EEO statement from the output text.\n",
    "\n",
    "# Create a new prompt template that includes the parser instructions\n",
    "parsed_job_posting_prompt = PromptTemplate(\n",
    "    input_variables=[\"job_title\", \"company\", \"location\", \"experience\"],\n",
    "    template=\"\"\"Create a job posting for a {job_title} position at {company} in {location}.\n",
    "    The candidate should have {experience} years of experience.\n",
    "    Follow these rules:\n",
    "    1. Start with a brief company description (2 sentences)\n",
    "    2. List 5 key responsibilities, each starting with an action verb\n",
    "    3. List 5 required qualifications, each in a single sentence\n",
    "    4. End with a standardized equal opportunity statement\n",
    "    \n",
    "    Format the output EXACTLY as follows:\n",
    "    COMPANY: [Company Description]\n",
    "    \n",
    "    RESPONSIBILITIES:\n",
    "    - [Responsibility 1]\n",
    "    - [Responsibility 2]\n",
    "    - [Responsibility 3]\n",
    "    - [Responsibility 4]\n",
    "    - [Responsibility 5]\n",
    "    \n",
    "    QUALIFICATIONS:\n",
    "    - [Qualification 1]\n",
    "    - [Qualification 2]\n",
    "    - [Qualification 3]\n",
    "    - [Qualification 4]\n",
    "    - [Qualification 5]\n",
    "    \n",
    "    EEO: [Equal Opportunity Statement]\n",
    "    \"\"\"\n",
    ")\n",
    "\n",
    "def clean_output(output):\n",
    "    for key, value in output.items():\n",
    "        if isinstance(value, str):\n",
    "            # Remove leading/trailing whitespace and normalize newlines\n",
    "            output[key] = re.sub(r'\\n\\s*', '\\n', value.strip())\n",
    "    return output\n",
    "\n",
    "# Generate the parsed output\n",
    "chain = parsed_job_posting_prompt | llm\n",
    "raw_output = chain.invoke(input_variables).content\n",
    "\n",
    "# Parse and clean the output\n",
    "parsed_output = regex_parser.parse(raw_output)\n",
    "cleaned_output = clean_output(parsed_output)\n",
    "\n",
    "# Display the parsed output\n",
    "print(\"Parsed Output:\")\n",
    "for key, value in cleaned_output.items():\n",
    "    print(f\"{key.upper()}:\")\n",
    "    print(value)\n",
    "    print()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Implementing Additional Constraints\n",
    "\n",
    "Let's create a more complex constrained generation task: generating a product review with specific criteria."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model Output:\n",
      "----------------------------------------\n",
      "Rating: 4 out of 5 stars\n",
      "\n",
      "Pros:\n",
      "1. The camera quality on Smartphone X is outstanding, capturing vibrant colors and sharp details even in low light. The multiple lens options provide versatility for different photography styles, making it a great choice for both casual users and photography enthusiasts.\n",
      "2. Battery life is impressive, lasting a full day with heavy usage. Quick charging capabilities ensure that you can get back to using your phone in no time, which is a huge plus for those on the go.\n",
      "3. The sleek design and lightweight build make Smartphone X comfortable to hold and use throughout the day. Its premium feel and modern aesthetics also make it visually appealing.\n",
      "\n",
      "Cons:\n",
      "1. Although the performance is generally smooth, there can be occasional lag when multitasking with resource-heavy applications. This might be a drawback for users who rely heavily on their devices for productivity.\n",
      "2. The lack of expandable storage is a limitation for those who need extra space for apps, photos, and videos. Users may find themselves needing to manage their storage more frequently as a result.\n",
      "\n",
      "Recommendation: Overall, Smartphone X is a fantastic choice for anyone seeking a powerful and stylish device.\n",
      "----------------------------------------\n",
      "\n"
     ]
    }
   ],
   "source": [
    "review_prompt = PromptTemplate(\n",
    "    input_variables=[\"product\", \"rating\", \"pros\", \"cons\", \"word_limit\"],\n",
    "    template=\"\"\"Write a product review for {product} with the following constraints:\n",
    "    1. The review should have a {rating}-star rating (out of 5)\n",
    "    2. Include exactly {pros} pros and {cons} cons\n",
    "    3. Use between 2 and 3 sentences for each pro and con\n",
    "    4. The entire review should be under {word_limit} words\n",
    "    5. End with a one-sentence recommendation\n",
    "    \n",
    "    Format the review as follows:\n",
    "    Rating: [X] out of 5 stars\n",
    "    \n",
    "    Pros:\n",
    "    1. [Pro 1]\n",
    "    2. [Pro 2]\n",
    "    ...\n",
    "    \n",
    "    Cons:\n",
    "    1. [Con 1]\n",
    "    2. [Con 2]\n",
    "    ...\n",
    "    \n",
    "    Recommendation: [One-sentence recommendation]\n",
    "    \"\"\"\n",
    ")\n",
    "\n",
    "# Generate the constrained review\n",
    "input_variables = {\n",
    "    \"product\": \"Smartphone X\",\n",
    "    \"rating\": \"4\",\n",
    "    \"pros\": \"3\",\n",
    "    \"cons\": \"2\",\n",
    "    \"word_limit\": \"200\"\n",
    "}\n",
    "\n",
    "chain = review_prompt | llm\n",
    "output = chain.invoke(input_variables).content\n",
    "display_output(output)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": ".venv",
   "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": "3.12.0"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}