How to use few-shot prompting with tool calling
This guide assumes familiarity with the following concepts:
For more complex tool use itβs very useful to add few-shot examples to
the prompt. We can do this by adding AIMessages
with ToolCalls
and
corresponding ToolMessages
to our prompt.
First define a model and a calculator tool:
import { tool } from "@langchain/core/tools";
import { z } from "zod";
import { ChatOpenAI } from "@langchain/openai";
const llm = new ChatOpenAI({ model: "gpt-4o", temperature: 0 });
/**
* Note that the descriptions here are crucial, as they will be passed along
* to the model along with the class name.
*/
const calculatorSchema = z.object({
operation: z
.enum(["add", "subtract", "multiply", "divide"])
.describe("The type of operation to execute."),
number1: z.number().describe("The first number to operate on."),
number2: z.number().describe("The second number to operate on."),
});
const calculatorTool = tool(
async ({ operation, number1, number2 }) => {
// Functions must return strings
if (operation === "add") {
return `${number1 + number2}`;
} else if (operation === "subtract") {
return `${number1 - number2}`;
} else if (operation === "multiply") {
return `${number1 * number2}`;
} else if (operation === "divide") {
return `${number1 / number2}`;
} else {
throw new Error("Invalid operation.");
}
},
{
name: "calculator",
description: "Can perform mathematical operations.",
schema: calculatorSchema,
}
);
const llmWithTools = llm.bindTools([calculatorTool]);
Our calculator can handle common addition, subtraction, multiplication,
and division. But what happens if we ask about a new mathematical
operator, π¦
?
Letβs see what happens when we use it naively:
const res = await llmWithTools.invoke("What is 3 π¦ 12");
console.log(res.content);
console.log(res.tool_calls);
[
{
name: 'calculator',
args: { operation: 'multiply', number1: 3, number2: 12 },
type: 'tool_call',
id: 'call_I0oQGmdESpIgcf91ej30p9aR'
}
]
It doesnβt quite know how to interpret π¦
as an operation, and it
defaults to multiply
. Now, letβs try giving it some examples in the
form of a manufactured messages to steer it towards divide
:
import { HumanMessage, AIMessage, ToolMessage } from "@langchain/core/messages";
const res = await llmWithTools.invoke([
new HumanMessage("What is 333382 π¦ 1932?"),
new AIMessage({
content:
"The π¦ operator is shorthand for division, so we call the divide tool.",
tool_calls: [
{
id: "12345",
name: "calculator",
args: {
number1: 333382,
number2: 1932,
operation: "divide",
},
},
],
}),
new ToolMessage({
tool_call_id: "12345",
content: "The answer is 172.558.",
}),
new AIMessage("The answer is 172.558."),
new HumanMessage("What is 6 π¦ 2?"),
new AIMessage({
content:
"The π¦ operator is shorthand for division, so we call the divide tool.",
tool_calls: [
{
id: "54321",
name: "calculator",
args: {
number1: 6,
number2: 2,
operation: "divide",
},
},
],
}),
new ToolMessage({
tool_call_id: "54321",
content: "The answer is 3.",
}),
new AIMessage("The answer is 3."),
new HumanMessage("What is 3 π¦ 12?"),
]);
console.log(res.tool_calls);
[
{
name: 'calculator',
args: { number1: 3, number2: 12, operation: 'divide' },
type: 'tool_call',
id: 'call_O6M4yDaA6s8oDqs2Zfl7TZAp'
}
]
And we can see that it now equates π¦
with the divide
operation in
the correct way!
Relatedβ
- Stream tool calls
- Pass runtime values to tools
- Getting structured outputs from models