Quick Start
Language models take text as input - that text is commonly referred to as a prompt. Typically this is not simply a hardcoded string but rather a combination of a template, some examples, and user input. LangChain provides several classes and functions to make constructing and working with prompts easy.
What is a prompt template?
A prompt template refers to a reproducible way to generate a prompt. It contains a text string ("the template"), that can take in a set of parameters from the end user and generates a prompt.
A prompt template can contain:
- instructions to the language model,
- a set of few shot examples to help the language model generate a better response,
- a question to the language model.
Here's a simple example:
- F-String
- Mustache
import { PromptTemplate } from "@langchain/core/prompts";
// If a template is passed in, the input variables are inferred automatically from the template.
const prompt = PromptTemplate.fromTemplate(
`You are a naming consultant for new companies.
What is a good name for a company that makes {product}?`
);
const formattedPrompt = await prompt.format({
product: "colorful socks",
});
/*
You are a naming consultant for new companies.
What is a good name for a company that makes colorful socks?
*/
API Reference:
- PromptTemplate from
@langchain/core/prompts
import { PromptTemplate } from "@langchain/core/prompts";
// If a template is passed in, the input variables are inferred automatically from the template.
const prompt = PromptTemplate.fromTemplate(
`You are a naming consultant for new companies.
What is a good name for a company that makes {{product}}?`,
{
templateFormat: "mustache",
}
);
const formattedPrompt = await prompt.format({
product: "colorful socks",
});
/*
You are a naming consultant for new companies.
What is a good name for a company that makes colorful socks?
*/
API Reference:
- PromptTemplate from
@langchain/core/prompts
Create a prompt template
You can create simple hardcoded prompts using the PromptTemplate
class. Prompt templates can take any number of input variables, and can be formatted to generate a prompt.
- F-String
- Mustache
import { PromptTemplate } from "@langchain/core/prompts";
// An example prompt with no input variables
const noInputPrompt = new PromptTemplate({
inputVariables: [],
template: "Tell me a joke.",
});
const formattedNoInputPrompt = await noInputPrompt.format({});
console.log(formattedNoInputPrompt);
// "Tell me a joke."
// An example prompt with one input variable
const oneInputPrompt = new PromptTemplate({
inputVariables: ["adjective"],
template: "Tell me a {adjective} joke.",
});
const formattedOneInputPrompt = await oneInputPrompt.format({
adjective: "funny",
});
console.log(formattedOneInputPrompt);
// "Tell me a funny joke."
// An example prompt with multiple input variables
const multipleInputPrompt = new PromptTemplate({
inputVariables: ["adjective", "content"],
template: "Tell me a {adjective} joke about {content}.",
});
const formattedMultipleInputPrompt = await multipleInputPrompt.format({
adjective: "funny",
content: "chickens",
});
console.log(formattedMultipleInputPrompt);
// "Tell me a funny joke about chickens."
API Reference:
- PromptTemplate from
@langchain/core/prompts
import { PromptTemplate } from "@langchain/core/prompts";
// An example prompt with no input variables
const noInputPrompt = new PromptTemplate({
inputVariables: [],
template: "Tell me a joke.",
});
const formattedNoInputPrompt = await noInputPrompt.format({});
console.log(formattedNoInputPrompt);
// "Tell me a joke."
// An example prompt with one input variable
const oneInputPrompt = new PromptTemplate({
inputVariables: ["adjective"],
template: "Tell me a {{adjective}} joke.",
templateFormat: "mustache",
});
const formattedOneInputPrompt = await oneInputPrompt.format({
adjective: "funny",
});
console.log(formattedOneInputPrompt);
// "Tell me a funny joke."
// An example prompt with multiple input variables
const multipleInputPrompt = new PromptTemplate({
inputVariables: ["adjective", "content"],
template: "Tell me a {{adjective}} joke about {{content}}.",
templateFormat: "mustache",
});
const formattedMultipleInputPrompt = await multipleInputPrompt.format({
adjective: "funny",
content: "chickens",
});
console.log(formattedMultipleInputPrompt);
// "Tell me a funny joke about chickens."
API Reference:
- PromptTemplate from
@langchain/core/prompts
If you do not wish to specify inputVariables
manually, you can also create a PromptTemplate
using the fromTemplate
class method. LangChain will automatically infer the inputVariables
based on the template
passed.
- F-String
- Mustache
import { PromptTemplate } from "@langchain/core/prompts";
const template = "Tell me a {adjective} joke about {content}.";
const promptTemplate = PromptTemplate.fromTemplate(template);
console.log(promptTemplate.inputVariables);
// ['adjective', 'content']
const formattedPromptTemplate = await promptTemplate.format({
adjective: "funny",
content: "chickens",
});
console.log(formattedPromptTemplate);
// "Tell me a funny joke about chickens."
API Reference:
- PromptTemplate from
@langchain/core/prompts
import { PromptTemplate } from "@langchain/core/prompts";
const template = "Tell me a {{adjective}} joke about {{content}}.";
const promptTemplate = PromptTemplate.fromTemplate(template, {
templateFormat: "mustache",
});
console.log(promptTemplate.inputVariables);
// ['adjective', 'content']
const formattedPromptTemplate = await promptTemplate.format({
adjective: "funny",
content: "chickens",
});
console.log(formattedPromptTemplate);
// "Tell me a funny joke about chickens."
API Reference:
- PromptTemplate from
@langchain/core/prompts
You can create custom prompt templates that format the prompt in any way you want.
Chat prompt template
Chat Models take a list of chat messages as input - this list is commonly referred to as a prompt
.
These chat messages differ from raw string (which you would pass into a LLM) in that every message is associated with a role
.
For example, in OpenAI Chat Completion API, a chat message can be associated with an AI, human or system role. The model is supposed to follow instruction from system chat message more closely.
LangChain provides several prompt templates to make constructing and working with prompts easily. You are encouraged to use these chat related prompt templates instead of PromptTemplate
when invoking chat models to fully explore the model's potential.
import {
ChatPromptTemplate,
PromptTemplate,
SystemMessagePromptTemplate,
AIMessagePromptTemplate,
HumanMessagePromptTemplate,
} from "@langchain/core/prompts";
import {
AIMessage,
HumanMessage,
SystemMessage,
} from "@langchain/core/messages";
To create a message template associated with a role, you would use the corresponding <ROLE>MessagePromptTemplate
.
For convenience, you can also declare message prompt templates as tuples. These will be coerced to the proper prompt template types:
const systemTemplate =
"You are a helpful assistant that translates {input_language} to {output_language}.";
const humanTemplate = "{text}";
const chatPrompt = ChatPromptTemplate.fromMessages([
["system", systemTemplate],
["human", humanTemplate],
]);
// Format the messages
const formattedChatPrompt = await chatPrompt.formatMessages({
input_language: "English",
output_language: "French",
text: "I love programming.",
});
console.log(formattedChatPrompt);
/*
[
SystemMessage {
content: 'You are a helpful assistant that translates English to French.'
},
HumanMessage {
content: 'I love programming.'
}
]
*/
You can also use ChatPromptTemplate
's .formatPromptValue()
method -- this returns a PromptValue
, which you can convert to a string or Message object,
depending on whether you want to use the formatted value as input to an LLM or chat model.
If you prefer to use the message classes, there is a fromTemplate
method exposed on these classes.
This is what it would look like:
const template =
"You are a helpful assistant that translates {input_language} to {output_language}.";
const systemMessagePrompt = SystemMessagePromptTemplate.fromTemplate(template);
const humanTemplate = "{text}";
const humanMessagePrompt =
HumanMessagePromptTemplate.fromTemplate(humanTemplate);
If you wanted to construct the MessagePromptTemplate
more directly, you could create a PromptTemplate externally and then pass it in, e.g.:
const prompt = new PromptTemplate({
template:
"You are a helpful assistant that translates {input_language} to {output_language}.",
inputVariables: ["input_language", "output_language"],
});
const systemMessagePrompt2 = new SystemMessagePromptTemplate({
prompt,
});
Note: If using TypeScript, you can add typing to prompts created with .fromMessages
by passing a type parameter like this:
const chatPrompt = ChatPromptTemplate.fromMessages<{
input_language: string;
output_language: string;
text: string;
}>([systemMessagePrompt, humanMessagePrompt]);
Multi-modal prompts
import { HumanMessage } from "@langchain/core/messages";
import { ChatPromptTemplate } from "@langchain/core/prompts";
import fs from "node:fs/promises";
const hotdogImage = await fs.readFile("hotdog.jpg");
const base64Image = hotdogImage.toString("base64");
const imageURL = "https://avatars.githubusercontent.com/u/126733545?s=200&v=4";
const langchainLogoMessage = new HumanMessage({
content: [
{
type: "image_url",
image_url: {
url: "{imageURL}",
detail: "high",
},
},
],
});
const base64ImageMessage = new HumanMessage({
content: [
{
type: "image_url",
image_url: "data:image/jpeg;base64,{base64Image}",
},
],
});
const multiModalPrompt = ChatPromptTemplate.fromMessages([
["system", "You have 20:20 vision! Describe the user's image."],
langchainLogoMessage,
base64ImageMessage,
]);
const formattedPrompt = await multiModalPrompt.invoke({
imageURL,
base64Image,
});
console.log(JSON.stringify(formattedPrompt, null, 2));
/**
{
"kwargs": {
"messages": [
{
"kwargs": {
"content": "You have 20:20 vision! Describe the user's image.",
}
},
{
"kwargs": {
"content": [
{
"type": "image_url",
"image_url": {
"url": "https://avatars.githubusercontent.com/u/126733545?s=200&v=4",
"detail": "high"
}
}
],
}
},
{
"kwargs": {
"content": [
{
"type": "image_url",
"image_url": "data:image/png;base64,/9j/4AAQSkZJRgABAQEBLAEsAAD/4Q..."
}
],
}
}
]
}
}
*/
API Reference:
- HumanMessage from
@langchain/core/messages
- ChatPromptTemplate from
@langchain/core/prompts
LangSmith will render your images inside traces!
See the LangSmith trace here
You can also pass multi-modal prompt templates inline:
import { ChatPromptTemplate } from "@langchain/core/prompts";
import fs from "node:fs/promises";
const hotdogImage = await fs.readFile("hotdog.jpg");
// Convert the image to base64
const base64Image = hotdogImage.toString("base64");
const imageURL = "https://avatars.githubusercontent.com/u/126733545?s=200&v=4";
const multiModalPrompt = ChatPromptTemplate.fromMessages([
["system", "You have 20:20 vision! Describe the user's image."],
[
"human",
[
{
type: "image_url",
image_url: {
url: "{imageURL}",
detail: "high",
},
},
{
type: "image_url",
image_url: "data:image/jpeg;base64,{base64Image}",
},
],
],
]);
const formattedPrompt = await multiModalPrompt.invoke({
imageURL,
base64Image,
});
console.log(JSON.stringify(formattedPrompt, null, 2));
/**
{
"kwargs": {
"messages": [
{
"kwargs": {
"content": "You have 20:20 vision! Describe the user's image.",
}
},
{
"kwargs": {
"content": [
{
"type": "image_url",
"image_url": {
"url": "https://avatars.githubusercontent.com/u/126733545?s=200&v=4",
"detail": "high"
}
},
{
"type": "image_url",
"image_url": {
"url": "data:image/jpeg;base64,/9j/4AAQSkZJRgABAQEBLAEsAAD/4QBWRX...",
}
}
],
}
}
]
}
}
*/
API Reference:
- ChatPromptTemplate from
@langchain/core/prompts
See the LangSmith trace here