Pydantic and Prompt Engineering: The Essentials for Validating Language Model Outputs
<p>Large language models (LLMs) output text in reaction to our prompts. While this provides a satisfactory interaction when chatting with them, it poses a challenge when we aim to incorporate these models into our applications. For a seamless integration, it is often crucial to have these responses in a structured format that can be effectively used and forwarded to subsequent processes. This is where <strong>Pydantic</strong> comes to the rescue, bridging the gap between the text responses from LLMs and the structured input our applications need. Pydantic enables us to define a desired data model that the output of LLM can be cast into. This way, we can ensure structured, validated, and consequently, more useful responses from our LLMs.</p>
<h1>Query your LLM with a prompt</h1>
<p>To get a response from our LLM using a prompt, we use the below function.</p>
<p><iframe frameborder="0" height="474" scrolling="no" src="https://medium.com/media/0c282ac5aa6f81b4f051d005d66817bf" title="query_llm.py" width="680"></iframe></p>
<p>Let’s call this function with the following prompt:</p>
<p><iframe frameborder="0" height="320" scrolling="no" src="https://medium.com/media/c334566244bbf200b5eecf5efa5b782b" title="get_llm_responses.py" width="680"></iframe></p>
<h1>A Dive into Pydantic Modeling</h1>
<p>Utilizing Pydantic models, let’s craft a structured output for our data. We define a <code>CityResponse</code> model that captures key details - city name, country it's in, population, and local currency. We then bundle these cities in a <code>Cities</code> model. This process helps us yield a custom listing of cities, with specifics and neatly packaged for insightful data handling. It is essential to supply a description for every field, as <a href="https://docs.langchain.com/docs/" rel="noopener ugc nofollow" target="_blank">LangChain</a> will utilize these for generating a prompt.</p>
<p><a href="https://medium.com/@azizbenothman76/pydantic-and-prompt-engineering-the-essentials-for-validating-language-model-outputs-e48553eb4a3b">Click Here</a></p>