📘 Overview

Text2MultiHopQAGenerator is a multi-hop question-answer pair generator operator designed to automatically produce questions and answers that require multi-step reasoning from a given text. This operator leverages a Large Language Model (LLM) to transform input text into a structured set of reasoning-based QA pairs. It is suitable for building complex QA datasets or evaluating a model’s reasoning ability.

init Function

__init__(self, llm_serving, seed=0, lang="en", prompt_template=None, num_q=5)

init Parameter Description

Parameter	Type	Default	Description
llm_serving	LLMServingABC	Required	The LLM service instance used for inference and generation.
seed	int	0	Random seed to ensure reproducibility of results.
lang	str	"en"	Specifies the output language, e.g., 'en' (English) or 'zh' (Chinese).
prompt_template	PromptABC	Text2MultiHopQAGeneratorPrompt()	Prompt template object for constructing model input.
num_q	int	5	The maximum number of QA pairs to generate for each input text.

Prompt Template Description

Prompt Template Name	Main Purpose	Application Scenario	Feature Description
Text2MultiHopQAGeneratorPrompt	Generate multi-hop QA pairs from text	Scenarios requiring complex reasoning questions from long context	Built-in specialized template that guides the model to generate questions, reasoning steps, final answers, and supporting facts to ensure structured and logical outputs.

run Function

run(self, storage, input_key='cleaned_chunk', output_key='QA_pairs', output_meta_key='QA_metadata')

Parameters

Name	Type	Default	Description
storage	DataFlowStorage	Required	Data flow storage instance responsible for reading and writing data.
input_key	str	"cleaned_chunk"	Input column name corresponding to the context text field.
output_key	str	"QA_pairs"	Output column name corresponding to the generated multi-hop QA pairs list.
output_meta_key	str	"QA_metadata"	Output metadata column name corresponding to the generated metadata information.

🧠 Example Usage

self.knowledge_cleaning_step4 = Text2MultiHopQAGenerator(
    llm_serving=self.llm_serving,
    lang="en",
    num_q = 5
)
self.knowledge_cleaning_step4.run(
    storage=self.storage.step(),
    # input_key=
    # output_key=
)

🧾 Default Output Format

Field	Type	Description
text	str	The processed original context text.
qa_pairs	List[Dict]	List of generated multi-hop QA pairs, each containing question, answer, reasoning steps, etc.
metadata	Dict	Metadata containing source, timestamp, complexity, and other information.

Example Input:

{
"cleaned_chunk":"Mona Lisa was painted by Leonardo da Vinci. Leonardo da Vinci was born in the Republic of Florence. The Republic of Florence was a state in what is now Italy."
}

Example Output:

{
    "cleaned_chunk": "Mona Lisa was painted by Leonardo da Vinci. Leonardo da Vinci was born in the Republic of Florence. The Republic of Florence was a state in what is now Italy.",
    "QA_pairs": [
        {
            "question": "In which modern country was the painter of the Mona Lisa born?",
            "reasoning_steps": [
                {
                    "step": "Identify the painter of the Mona Lisa, which is Leonardo da Vinci."
                },
                {
                    "step": "Find the birthplace of Leonardo da Vinci, which is the Republic of Florence."
                },
                {
                    "step": "Determine the modern-day location of the Republic of Florence, which is Italy."
                }
            ],
            "answer": "Italy",
            "supporting_facts": [
                "Mona Lisa was painted by Leonardo da Vinci.",
                "Leonardo da Vinci was born in the Republic of Florence.",
                "The Republic of Florence was a state in what is now Italy."
            ],
            "type": "History"
        }
    ],
    "QA_metadata": {
        "source": "default_source",
        "timestamp": "2023-10-27T10:00:00Z",
        "complexity": 3
    }
}