run-llama/llama_index · error · ValueError

Prompt should have output parser.

Error message

Prompt should have output parser.

What it means

LLMQuestionGenerator parses the LLM's sub-question output with prompt.output_parser. The constructor requires the prompt template to carry an output parser (a SubQuestionOutputParser) so parsed text can become SubQuestion objects; without one it raises ValueError('Prompt should have output parser.')

Source

Thrown at llama-index-core/llama_index/core/question_gen/llm_generators.py:30

)
from llama_index.core.question_gen.types import BaseQuestionGenerator, SubQuestion
from llama_index.core.schema import QueryBundle
from llama_index.core.settings import Settings
from llama_index.core.tools.types import ToolMetadata
from llama_index.core.types import BaseOutputParser


class LLMQuestionGenerator(BaseQuestionGenerator):
    def __init__(
        self,
        llm: LLM,
        prompt: BasePromptTemplate,
    ) -> None:
        self._llm = llm
        self._prompt = prompt

        if self._prompt.output_parser is None:
            raise ValueError("Prompt should have output parser.")

    @classmethod
    def from_defaults(
        cls,
        llm: Optional[LLM] = None,
        prompt_template_str: Optional[str] = None,
        output_parser: Optional[BaseOutputParser] = None,
    ) -> "LLMQuestionGenerator":
        # optionally initialize defaults
        llm = llm or Settings.llm
        prompt_template_str = prompt_template_str or DEFAULT_SUB_QUESTION_PROMPT_TMPL
        output_parser = output_parser or SubQuestionOutputParser()

        # construct prompt
        prompt = PromptTemplate(
            template=prompt_template_str,
            output_parser=output_parser,
            prompt_type=PromptType.SUB_QUESTION,

View on GitHub (pinned to afd0fef371)

Solutions

1. Attach the parser when building a custom prompt: PromptTemplate(template, output_parser=SubQuestionOutputParser())
2. Or prefer from_defaults(prompt_template_str=...) which wires SubQuestionOutputParser automatically
3. Keep the {output_cls} / JSON structure tokens in a custom template so the parser still works

Example fix

// before
from llama_index.core import PromptTemplate
prompt = PromptTemplate(custom_sub_question_template)  # no parser
gen = LLMQuestionGenerator(llm=llm, prompt=prompt)

// after
from llama_index.core.question_gen.output_parser import SubQuestionOutputParser
prompt = PromptTemplate(custom_sub_question_template, output_parser=SubQuestionOutputParser())
gen = LLMQuestionGenerator(llm=llm, prompt=prompt)

from llama_index.core.question_gen.output_parser import SubQuestionOutputParser

def make_question_prompt(template_str: str):
    prompt = PromptTemplate(template_str)
    if prompt.output_parser is None:
        prompt = PromptTemplate(template_str, output_parser=SubQuestionOutputParser())
    return prompt

def prompt_has_output_parser(prompt) -> bool:
    """LLMQuestionGenerator requires a parsed output."""
    return prompt.output_parser is not None

When it happens

Trigger: Instantiating LLMQuestionGenerator(llm=..., prompt=PromptTemplate(...)) with a plain PromptTemplate that has no output_parser, or from_defaults(prompt_template_str=custom_str, output_parser=None) is fine (it defaults) but a custom prompt object lacking a parser is not.

Common situations: Swapping in a custom sub-question prompt by constructing BasePromptTemplate/PromptTemplate directly without attaching SubQuestionOutputParser.

Related errors

• All agents must have a name in a multi-agent workflow
• All agents must have a description in a multi-agent workflow
• Initial state is not supported per-agent in AgentWorkflow
• Exactly one root agent must be provided
• Root agent {root_agent} not found in provided agents

AI-assisted analysis of run-llama/llama_index@afd0fef371 (2026-08-15). Data as JSON: /api/errors/c1a00fd1e049af91. Report an issue: GitHub.