AI Paper Atlas

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

SFT Doesn’t Always Hurt General Capabilities: Revisiting Domain-Specific Fine-Tuning in LLMs

Jiacheng Lin, Zhongruo Wang, Kun Qian, Tian Wang, Arvind Srinivasan, Hansi Zeng, Ruochen Jiao, Xie Zhou, Jiri Gesi, Dakuo Wang, Yufan Guo, Kai Zhong and 5 more

ICLR 2026 · Large language models › Instruction and fine-tuning · approach: Efficient LLM fine-tuning systems · 18 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
use smaller learning rate 1e-6 instead of 5e-6 (ablation)Standard SFT · General capability avg (MedCalc)average score: 0.3805 -> 0.6924Improved
use smaller learning rate 1e-6 instead of 5e-6 (ablation)Standard SFT · MedCalcaccuracy: 0.5583 -> 0.5339No clear effect
add TALR at lr 5e-6Standard SFT · General capability avg (MedCalc, lr 5e-6)average score: 0.3805 -> 0.5018Improved
add TALR at lr 5e-6Standard SFT · MedCalc domain (lr 5e-6)accuracy: 0.5583 -> 0.5417Hurt
add TALR at lr 1e-6Standard SFT · General capability avg (MedCalc, lr 1e-6)average score: 0.6924 -> 0.7169Improved
use LoRA instead of full fine-tuningStandard SFT · MedCalc domain (lr 1e-6)accuracy: 0.5339 -> 0.1813Hurt
add Wise-FT model averagingStandard SFT · MedCalc domain (lr 1e-6)accuracy: 0.5339 -> 0.1983Hurt
add KL regularization (coefficient 0.001) (ablation)Standard SFT · MedCalc (Qwen2.5-3B-Instruct)general performanceNo clear effect
label-only supervision (no CoT) on ESCI (ablation)SFT · ESCItrade-off across learning ratesImproved
lower learning rate to 5e-7 on MetaMathQA (ablation)SFT DeepSeek-Coder-7B · GSM8Kdomain accuracyHurt

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