Prompt templates en LangChain

What is a "prompt template" and how is it used?

Getting started with language models may seem complicated, but prompt templates greatly simplify this process. A prompt template is a structure that is used to convert user input into a clear instruction, thus guiding the desired output of the language model. This approach is crucial for those seeking to get the most out of tools such aslarge language models or chat models.

Prompt templates use a dictionary as input where each key represents a variable to be filled in. The output generated by a template, known as a prompt value, can be used in various ways, either as a text string or a list of messages.

How is a string prompt template created?

One of the simplest types of prompt templates is the string prompt template. Here, you work with a simple input and format it as a text string.

1. Define the template: You define it using the PromptTemplate module of langchain.core.
2. Instruction to the template: Example, "tell me a joke".
3. Input variable: The key within the dictionary, e.g., { topic }, where the topic of the joke is specified.

This approach allows a flexible and clear way to interact with language models in multiple languages.

from langchain.core.prompts import PromptTemplate
prompt_template = PromptTemplate(text="Tell me a joke about {topic}.")result = prompt_template.invoke({"topic": "cats"})

How do you turn templates into a chat conversation?

In addition to the string approach, prompt templates can also be structured as lists of messages, thus creating a conversational dynamic.

1. Create a basic conversation: Using the ChatPromptTemplate to give a role to each party, either System or User.
2. Explicitly define roles: Although from the system or in written form, the roles involved must be defined.
3. Send instructions: Example, "you are a helpful assistant", "tell me a joke about cats".

Using ChatPromptTemplate provides a framework to integrate multiple interactions, enriching the dialog.

from langchain.core.prompts import ChatPromptTemplate
chat_prompt_template = ChatPromptTemplate(messages=[ {"system": "You're a helpful assistant."}, {"user": "Tell me a joke about {topic}."}])result = chat_prompt_template.invoke({"topic": "cats"})

What is "Message Placeholder" and how is it integrated?

The Message Placeholder is essential for those who need more flexibility when integrating multiple messages into a dialog.

1. Use of placeholders: They allow you to designate specific points in the template where a list of messages can be added.
2. Tailoring the conversation: Ideal for setting up complex dialogs where it is necessary to vary the participants or go into more detail.

This technique is particularly useful when additional data arises or messages need to be placed in already established sequences.

from langchain.core.prompts import ChatPromptTemplate, HumanMessage, MessagePlaceholder
prompt_template = ChatPromptTemplate(messages=[ {"system": "You are a helpful assistant."}, MessagePlaceholder()])
new_messages = [ HumanMessage(content="Hello"), HumanMessage(content="Goodbye")]
result = prompt_template.invoke({"messages":  new_messages})

The use of prompt templates not only improves the clarity of instructions, but also allows for more involved and enriching interactions with language models. We encourage you to experiment with these tools and share your achievements to continue learning together.