Implementation:Datajuicer Data juicer OptimizeQueryMapper

Knowledge Sources	Datajuicer_Data_juicer
Domains	Data_Processing, Mapping
Last Updated	2026-02-14 16:00 GMT

Overview

Concrete tool for optimizing queries in question-answer pairs provided by Data-Juicer.

Description

OptimizeQueryMapper is a mapper operator that extends OptimizeQAMapper to optimize only the query (question) in a QA pair, making it more specific and detailed while ensuring the original answer can still address it. It uses a specialized Chinese system prompt instructing the model to optimize only the question. The parse_output method returns the stripped raw output as the optimized query and None for the answer, leaving the original answer unchanged. Inherits all model configuration, retry logic, and VLLM/HF/API support from the parent class. Requires CUDA acceleration.

Usage

Use when answer quality is already adequate but questions need to be more specific or detailed for better instruction-following training data.

Code Reference

Source Location

• Repository: Datajuicer_Data_juicer
• File: data_juicer/ops/mapper/optimize_query_mapper.py

Signature

@OPERATORS.register_module("optimize_query_mapper")
class OptimizeQueryMapper(OptimizeQAMapper):
    def __init__(self,
                 api_or_hf_model: str = "Qwen/Qwen2.5-7B-Instruct",
                 is_hf_model: bool = True,
                 *,
                 api_endpoint: Optional[str] = None,
                 response_path: Optional[str] = None,
                 system_prompt: Optional[str] = None,
                 input_template: Optional[str] = None,
                 qa_pair_template: Optional[str] = None,
                 output_pattern: Optional[str] = None,
                 try_num: PositiveInt = 3,
                 enable_vllm: bool = False,
                 model_params: Optional[Dict] = None,
                 sampling_params: Optional[Dict] = None,
                 **kwargs):

Import

from data_juicer.ops.mapper.optimize_query_mapper import OptimizeQueryMapper

I/O Contract

Inputs

Name	Type	Required	Description
api_or_hf_model	str	No	API or HuggingFace model name, defaults to "Qwen/Qwen2.5-7B-Instruct"
is_hf_model	bool	No	If True, use HuggingFace model; otherwise use API, defaults to True
api_endpoint	Optional[str]	No	URL endpoint for the API
response_path	Optional[str]	No	Path to extract content from API response
system_prompt	Optional[str]	No	System prompt for guiding optimization; defaults to Chinese query optimization prompt
input_template	Optional[str]	No	Template for building model input
qa_pair_template	Optional[str]	No	Template for formatting question-answer pairs
output_pattern	Optional[str]	No	Regex pattern to extract question and answer from model response
try_num	PositiveInt	No	Number of retry attempts on error, defaults to 3
enable_vllm	bool	No	Whether to use VLLM for inference acceleration, defaults to False
model_params	Optional[Dict]	No	Parameters for initializing the model
sampling_params	Optional[Dict]	No	Sampling parameters for text generation

Outputs

Name	Type	Description
samples	Dict	Transformed samples with optimized query replacing the original question

Usage Examples

process:
  - optimize_query_mapper:
      api_or_hf_model: "Qwen/Qwen2.5-7B-Instruct"
      is_hf_model: true
      try_num: 3

Related Pages

• Environment:Datajuicer_Data_juicer_Python_Runtime_Environment