AI Paper Atlas

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

MMKU-Bench: A Multimodal Update Benchmark for Diverse Visual Knowledge

Baochen Fu, Yuntao Du, Cheng Chang, Baihao Jin, Wenzhi Deng, Muhao Xu, Hongmei Yan, Weiye Song, Yi Wan

ICML 2026 · Continual learning › Continual fine-tuning of LLMs · approach: Mixture-of-experts for multimodal continual tuning · 0 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
provide golden context via RAGInternVL2.5-8B · MMKU-Bench Updated vs UnknownCorrect (CEM): Updated 67.02 vs Unknown 76.03Mixed
use RLHF (CPO/SimPO) instead of SFTInternVL2.5-8B · MMKU-Bench UpdatedOutdated (CEM): Full-FT 5.53 -> SimPO 3.94Improved
use RLHF instead of SFT for overall injectionInternVL2.5-8B · MMKU-Bench UpdatedCorrect (CEM): Full-FT 35.19 -> CPO 33.35No clear effect
use knowledge editing (WISE) for large-scale injectionInternVL2.5-8B · MMKU-Bench UpdatedCorrect (CEM): 35.19 -> 6.06Hurt
use GRACE editingInternVL2.5-8B · MMKU-Bench UpdatedCorrect (CEM): 0.61Hurt
inject updated instead of unknown knowledge with Full-FT (ablation)InternVL2.5-8B · MMEscore: Unknown 1090.6 vs Updated 499.1Hurt
inject updated instead of unknown knowledge with Full-FTQwen2.5-VL-7B · MMEscore: Unknown 1162.7 vs Updated 393.9Hurt
use LoRA instead of full fine-tuning (ablation)InternVL2.5-8B · MME (Unknown)score: Full-FT 1090.6 -> LoRA 1515.0Improved
use WISE editing to preserve capabilityInternVL2.5-8B · MMEscore: Vanilla 1688.2 vs WISE 1672.6Improved
fine-tune Qwen2.5-VL with LoRA on updated knowledgeQwen2.5-VL-7B · MMBenchaccuracy: 82.2 -> 9.1Hurt
match injection and evaluation modality (M->M)SFT/RLHF models · MMKU-Bench cross-modalCorrectImproved
inject multimodal but evaluate on text (M->T)SFT/RLHF models · MMKU-Bench UpdatedCorrectHurt

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