AI Paper Atlas

Research map / Continual learning / Continual fine-tuning of LLMs

MetaGDPO: Alleviating Catastrophic Forgetting with Metacognitive Knowledge Through Group Direct Preference Optimization

Lanxue Zhang, Yuqiang Xie, Fang Fang, Fanglong Dong, Rui Liu, Yanan Cao

AAAI 2026 · Continual learning › Continual fine-tuning of LLMs · approach: Continual pretraining of language models · 2 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
fine-tune with MetaGDPO instead of no fine-tuningQwen3-8B · Math avgaccuracy: 79.16 -> 80.08Improved
fine-tune with MetaGDPO instead of no fine-tuningQwen3-8B · General avgaccuracy: 72.09 -> 76.79Improved
fine-tune with MetaGDPO instead of no fine-tuningR1-Llama-8B · General avgaccuracy: 53.51 -> 58.34Improved
fine-tune with LIMO instead of no fine-tuningQwen3-8B · Math avgaccuracy: 79.16 -> 70.59Hurt
fine-tune with L+S instead of no fine-tuningR1-Qwen-7B · GSM8Kaccuracy: 89.69 -> 46.47Hurt
fine-tune with STAR-1 instead of no fine-tuningQwen3-8B · General avgaccuracy: 72.09 -> 41.46Hurt
fine-tune with STAR-1 instead of no fine-tuningQwen3-8B · Safety avgsafety ratio: 83.87 -> 96.39Improved
add advantage weights to GDPO (ablation)GDPO · overall averageaverage score: 80.25 (without advantages)Improved
increase group size G from 2 to 10 (ablation)GDPO · overall performanceaverage scoreImproved
train on only 10 instances (ablation)GDPO · overall performanceaverage scoreImproved
use GDPO instead of SFT (ablation)Qwen3-8B · avg across math/commonsense/safetyaverage score: nearly 17%Improved
use GDPO instead of DPO (ablation)Qwen3-8B · avg across math/commonsense/safetyaverage score: nearly 3%Improved

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