📘 Overview

SFTGeneratorSeed is a supervised fine-tuning (SFT) data generation operator. Based on given document content, it calls a large language model (LLM) to generate instruction-response data pairs in supervised fine-tuning format. The operator supports users to control the specific requirements of generated content through custom prompts, enabling automated construction of high-quality SFT datasets from raw documents.

__init__

def __init__(self, llm_serving: LLMServingABC, custom_prompt: str)

init Parameters

Parameter	Type	Default	Description
llm_serving	LLMServingABC	Required	The Large Language Model serving instance, used to execute inference and generation.
custom_prompt	str	Required	User-defined prompt used to guide the model to generate question-answer pairs of specific style or content.

Prompt Template Descriptions

Prompt Template Name	Primary Use	Applicable Scenarios	Feature Description

run

def run(self, storage: DataFlowStorage, input_key: str = "raw_content")

Executes the main operator logic, reads a DataFrame containing raw documents from storage, generates instructions and answers for each document content, and writes the results back to storage.

Parameters

Name	Type	Default	Description
storage	DataFlowStorage	None	The DataFlowStorage instance, responsible for reading and writing data.
input_key	str	"raw_content"	Input column name, corresponding to the field containing raw document content.

🧠 Example Usage

🧾 Default Output Format

Field	Type	Description
instruction	str	The instruction or question generated by the model based on the original document.
output	str	The answer given by the model for the generated instruction.
raw_content	str	The original document content used to generate the question-answer pair.

Example Input:

{
"raw_content": "Penicillin, also known as penicillium, is a type of antibiotic that contains penicillanic acid in its molecular structure, which can destroy the cell walls of bacteria and act as a bactericide during the bacterial cell reproduction period. In 1928, British bacteriologist Alexander Fleming discovered in the laboratory that a type of mold (penicillium) could secrete a substance to inhibit the growth of staphylococcus. He named this substance penicillin."
}

Example Output:

{
"instruction": "Who discovered penicillin in 1928, and what is its main function?",
"output": "Penicillin was discovered by British bacteriologist Alexander Fleming in 1928. Its main function is to destroy the cell walls of bacteria and act as a bactericide during the bacterial reproduction period.",
"raw_content": "Penicillin, also known as penicillium, is a type of antibiotic that contains penicillanic acid in its molecular structure, which can destroy the cell walls of bacteria and act as a bactericide during the bacterial cell reproduction period. In 1928, British bacteriologist Alexander Fleming discovered in the laboratory that a type of mold (penicillium) could secrete a substance to inhibit the growth of staphylococcus. He named this substance penicillin."
}