📘 Overview

The SQLGenerator is an operator designed to synthesize SQL queries based on provided database information. It interacts with a database manager to retrieve schema details and uses a large language model (LLM) to generate a specified number of SQL queries, covering various complexities and styles. This is particularly useful for creating datasets for Text-to-SQL tasks.

__init__ function

def __init__(self,
             llm_serving: LLMServingABC,
             database_manager: DatabaseManager,
             generate_num: int = 300,
             prompt_template = None
    ):

init Parameters

Parameter	Type	Default Value	Description
llm_serving	LLMServingABC	Required	An instance of a large language model serving class, used for generating the SQL queries.
database_manager	DatabaseManager	Required	A manager instance that handles database connections and retrieves schema information (e.g., CREATE and INSERT statements).
generate_num	int	300	The number of SQL queries to generate for each database schema.
prompt_template	PromptABC	SelectSQLGeneratorPrompt()	The prompt template object used to construct the input for the LLM. Defaults to a standard SQL generation prompt if not provided.

Prompt Template Descriptions

Prompt Template Name	Primary Use	Applicable Scenarios	Feature Description

run function

def run(self,
        storage: DataFlowStorage,
        output_sql_key: str = "sql",
        output_db_id_key: str = "db_id"
    )

Parameters

Name	Type	Default Value	Description
storage	DataFlowStorage	Required	The DataFlow storage instance used for writing the generated data.
output_sql_key	str	"sql"	The column name in the output for the generated SQL query.
output_db_id_key	str	"db_id"	The column name in the output for the corresponding database ID.

🧠 Example Usage

🧾 Default Output Format

The operator processes database schemas provided by the DatabaseManager and writes a DataFrame to storage. Each row in the output corresponds to one generated SQL query.

Field	Type	Description
db_id	str	The identifier of the database for which the SQL was generated.
sql	str	The generated SQL query.

Example Input:

This operator does not process row-by-row JSON input. Instead, it uses the DatabaseManager to access database schemas.

Example Output:

{
"db_id": "concert_singer",
"sql": "SELECT name, country, age FROM singer ORDER BY age DESC"
}