📘 Overview

The ScenarioExpander is an operator designed to generate new or alternative scenarios based on original input scenarios. It utilizes a large language model (LLM) service to rewrite or reimagine the original content, creating different versions or expansions of the initial situation.

init

def __init__(self, llm_serving: LLMServingABC)

Parameter	Type	Default	Description
llm_serving	LLMServingABC	Required	LLM serving object implementing the LLMServingABC interface.

Prompt Template Descriptions

Prompt Template Name	Primary Use	Applicable Scenarios	Feature Description

run

def run(self, storage: DataFlowStorage, input_scenario_key: str, output_key: str = "modified_scenario")

Parameter	Type	Default	Description
storage	DataFlowStorage	Required	DataFlow storage instance, responsible for reading and writing data.
input_scenario_key	str	Required	Field name for the original scenario in the input data.
output_key	str	"modified_scenario"	Field name for the new scenario in the output data.

🧠 Example Usage

from dataflow.operators.conversations import ScenarioExpandGenerator
from dataflow.utils.storage import FileStorage
from dataflow.serving import APILLMServing_request
from dataflow.core import LLMServingABC

class ScenarioExpandGeneratorExample:
    def __init__(self, llm_serving: LLMServingABC = None):
        self.storage = FileStorage(
            first_entry_file_name="input.jsonl",
            cache_path="./cache_local",
            file_name_prefix="dataflow_cache_step",
            cache_type="jsonl",
        )

        self.llm_serving = APILLMServing_request(
            api_url="",
            model_name="gpt-4o",
            max_workers=30
        )

        self.generator = ScenarioExpandGenerator(
            llm_serving=self.llm_serving
        )

    def forward(self):
        self.generator.run(
            storage=self.storage.step(),
            input_scenario_key="original_scenario",
            output_key="modified_scenario"
        )

if __name__ == "__main__":
    pl = ScenarioExpandGeneratorExample()
    pl.forward()

🧾 Default Output Format

Field	Type	Description
[input_scenario_key]	str	The original input scenario text.
modified_scenario	str	The new scenario generated by the LLM.

Example Input:

{
"original_scenario":"A user is trying to log into their bank account but has forgotten their password."
}

Example Output:

{
"original_scenario":"A user is trying to log into their bank account but has forgotten their password.",
"modified_scenario":"A traveling salesperson needs to access their corporate expense report system from a hotel with unreliable Wi-Fi, and their two-factor authentication token has just expired."
}