Should .withStructuredOutput throw an error when get some wrong input?

Product Help LangChain
1

I started learn how to build app with LangChain recently.It’s convenience and powerful.

But it is really make me confused when i use model.withStructuredOutput(scheme) api. In my opinion, I expect a json output when i use this api.BUT IN FACT, it is often throw an OUTPUT_PARSING_FAILURE error.

For example usage:

import { ChatOllama, } from "@langchain/ollama";
import { z } from 'zod'

const llm = new ChatOllama({
  model: 'qwen3:32b'
})

const joke = z.object({
  setup: z.string().describe("The setup of the joke"),
  punchline: z.string().describe("The punchline to the joke"),
  rating: z.number().optional().describe("How funny the joke is, from 1 to 10"),
});

const structuredLlm = llm.withStructuredOutput(joke);

const res = await structuredLlm.invoke("abcd123456789qweasdgh");

console.log(res)

This will throw an error because i input something wrong on purpose. But in tutorial when i input the “right” message, it will also throw an error in OUTPUT_PARSING_FAILURE. Yeah you can say it’s possible a question about llm’self ability.But I think throw an error reduce the usability of withStructuredOutput(scheme) api, because user input is not under control.

To fix this, I have to turn to use prompt + parser + raw invoke. It’s helpful but not convinence like this api.

I’m wonder it’s neccesary to throw an error when structuredLLM gei somethin wrong? I think this might reduce robustness of the system.

Thankyou for reading my message!:smiley:

2

I figure out how to fix this, I add method: 'json_mode' when call withStructuredOutput :rofl: , forgive my ignorance