AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

SRFT: A Single-Stage Method with Supervised and Reinforcement Fine-Tuning for Reasoning

Yuqian Fu, Tinghong Chen, Jiajun Chai, Xihuai Wang, Songjun Tu, Guojun Yin, Wei Lin, Qichao Zhang, Yuanheng Zhu, Dongbin Zhao

ICLR 2026 · Large language models › Instruction and fine-tuning · approach: Fine-tuning for reasoning data · 76 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Add entropy-aware SFT weight w_SFT (instead of fixed 1.0) (ablation)SRFT · Math reasoning avg (5 benchmarks)avg accuracy: 55.1 -> 59.1Improved
Add entropy-aware RL weight w_RL (instead of fixed 1.0) (ablation)SRFT · Math reasoning avg (5 benchmarks)avg accuracy: 56.2 -> 59.1Improved
Apply RL then SFT instead of SFT then RLQwen2.5-Math-7B · Math reasoning avg (5 benchmarks)avg accuracy: 52.5 -> 37.4Hurt
Add KL constraint to RL->SFTRL->SFT · Math reasoning avg (5 benchmarks)avg accuracy: 37.4 -> 38.3No clear effect
Run SFT then RL instead of RL aloneQwen2.5-Math-7B · Math reasoning avg (5 benchmarks)avg accuracy: 49.4 -> 52.5Improved
Run SFT then RL instead of SFT aloneQwen2.5-Math-7B · Math reasoning avg (5 benchmarks)avg accuracy: 47.3 -> 52.5Improved
Use single-stage SRFT instead of two-stage SFT->RLQwen2.5-Math-7B · Math reasoning avg (5 benchmarks)avg accuracy: 54.5 -> 59.5Improved
Use single-stage SRFT instead of best SFT+RL baseline on OODQwen2.5-Math-7B · OOD avgavg accuracy: 57.8 -> 62.5Improved
Use SRFT instead of the best zero-RL baselineQwen2.5-Math-7B · Math reasoning avg (5 benchmarks)avg accuracy: +9.0 pointsImproved
Use SRFT instead of SFTQwen2.5-Math-7B · Math reasoning avg (5 benchmarks)avg accuracy: +4.8 pointsImproved
Use SRFT instead of SFT+RL methodsQwen2.5-Math-7B · Math reasoning avg (5 benchmarks)avg accuracy: +3.4 pointsImproved
Use the SRFT chat template instead of the Qwen2.5-Math template (ablation)Qwen2.5-Math-7B base · Math reasoning avgAverage: 0.218 -> 0.340Improved

Later papers that compared with this method

1 later accepted paper reports SRFT as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy