Welcome to the KLMS Tool version of TableSage

TableSage is a profiling tool for tabular data leveraging the powerful capabilities of LLMs, appropriate for data scientists.

Instructions

Input

TableSage can be executed from the KLMS with the following input json:

{
	"input": {
		"data": [
		    "XXXXXXXX-bucket/temp.csv"
		]
	},
	"output": {
		"profile": "/path/to/write/the/file"
    },
	"parameters": {
        "model": "llama3-8b-8192"
    },
    "secrets": {
        "endpoint": "Endpoint/of/LLM"
		"token": "Token/for/Endpoint"
	}
}

Input

• data (str, required)
  Path to the csv file.

Parameters

• model (str, required)
  Specifies the language model to use for processing. At least one model must be provided.

• table_prompt_ids (list, optional, default=None)
  List of prompt identifiers associated with table-level descriptions. If not provided, random table-specific prompts are used.

• column_prompt_ids (list, optional, default=None)
  List of prompt identifiers associated with column-level descriptions. If not provided, random column-specific prompts are used.

• official_table_description (str, optional, default=None)
  A predefined official description of the table. If provided, it is included in the prompt for additional context.

• official_column_descriptions (dict, optional, default=None)
  A mapping of column names to their official descriptions. If provided, these descriptions are incorporated into the prompt.

• no_prompts (int, optional, default=3)
  Controls the number of prompt variations to use in inference if table_prompt_ids=None or column_prompt_ids=None. If not specified, the default is 3.

Output

The output of TableSage has the following format:

{
    "message": "Profiler executed successfully!",
	"output": {
		"profile": "path_of_profile_file",
    }
	"metrics": {	
        "time": 1.698,
    },
	"status": "success"
}