Synthetic Data RL: Task Definition Is All You Need

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Guo, Yiduo, Guo, Zhen, Huang, Chuanwei, Wang, Zi-Ang, Zhang, Zekai, Yu, Haofei, Zhang, Huishuai, Shen, Yikang
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915300077731840
author Guo, Yiduo
Guo, Zhen
Huang, Chuanwei
Wang, Zi-Ang
Zhang, Zekai
Yu, Haofei
Zhang, Huishuai
Shen, Yikang
author_facet Guo, Yiduo
Guo, Zhen
Huang, Chuanwei
Wang, Zi-Ang
Zhang, Zekai
Yu, Haofei
Zhang, Huishuai
Shen, Yikang
contents Reinforcement learning (RL) is a powerful way to adapt foundation models to specialized tasks, but its reliance on large-scale human-labeled data limits broad adoption. We introduce Synthetic Data RL, a simple and general framework that reinforcement fine-tunes models using only synthetic data generated from a task definition. Our method first generates question and answer pairs from the task definition and retrieved documents, then adapts the difficulty of the question based on model solvability, and selects questions using the average pass rate of the model across samples for RL training. On Qwen-2.5-7B, our method achieves a 29.2% absolute improvement over the base model on GSM8K (+2.9 pp vs. instruction-tuned, +6.6 pp vs. Self-Instruct), 8.7% on MATH, 13.1% on GPQA (+7.0 pp vs. SynthLLM), 8.9% on MedQA, 17.7% on CQA (law) and 13.7% on CFA (finance). It surpasses supervised fine-tuning under the same data budget and nearly matches RL with full human data across datasets (e.g., +17.2 pp on GSM8K). Adding 100 human demonstrations improves the performance of GSM8K only by 0.4 pp, showing a limited added value. By reducing human data annotation, Synthetic Data RL enables scalable and efficient RL-based model adaptation. Code and demos are available at https://github.com/gydpku/Data_Synthesis_RL/.
format Preprint
id arxiv_https___arxiv_org_abs_2505_17063
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Synthetic Data RL: Task Definition Is All You Need
Guo, Yiduo
Guo, Zhen
Huang, Chuanwei
Wang, Zi-Ang
Zhang, Zekai
Yu, Haofei
Zhang, Huishuai
Shen, Yikang
Computation and Language
Artificial Intelligence
Machine Learning
Reinforcement learning (RL) is a powerful way to adapt foundation models to specialized tasks, but its reliance on large-scale human-labeled data limits broad adoption. We introduce Synthetic Data RL, a simple and general framework that reinforcement fine-tunes models using only synthetic data generated from a task definition. Our method first generates question and answer pairs from the task definition and retrieved documents, then adapts the difficulty of the question based on model solvability, and selects questions using the average pass rate of the model across samples for RL training. On Qwen-2.5-7B, our method achieves a 29.2% absolute improvement over the base model on GSM8K (+2.9 pp vs. instruction-tuned, +6.6 pp vs. Self-Instruct), 8.7% on MATH, 13.1% on GPQA (+7.0 pp vs. SynthLLM), 8.9% on MedQA, 17.7% on CQA (law) and 13.7% on CFA (finance). It surpasses supervised fine-tuning under the same data budget and nearly matches RL with full human data across datasets (e.g., +17.2 pp on GSM8K). Adding 100 human demonstrations improves the performance of GSM8K only by 0.4 pp, showing a limited added value. By reducing human data annotation, Synthetic Data RL enables scalable and efficient RL-based model adaptation. Code and demos are available at https://github.com/gydpku/Data_Synthesis_RL/.
title Synthetic Data RL: Task Definition Is All You Need
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2505.17063