AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

VPGTrans: Transfer Visual Prompt Generator across LLMs

Ao Zhang, Hao Fei, Yuan Yao, Wei Ji, Li Li, Zhiyuan Liu, Tat-Seng Chua

NeurIPS 2023 · Large language models › Language, safety and interpretability · approach: LLM unlearning and benchmarks · 5 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
Initialize projector with word-embedding converter before warm-up (ablation)linear projector warm-up OPT125M->1.3B · COCO captionCIDEr: 126.1 -> 130.2Improved
Use 5x learning rate for 1 epoch of projector warm-up (ablation)projector warm-up · COCO captionCIDEr: 133.1 vs 132.8 (3 epochs w/o init)Improved
Add projector warm-up before VPG tuning (stage-1) (ablation)VPG inheritance · VQAv2accuracyImproved
Inherit VPG trained on OPT125M instead of training from scratchOPT350M · COCO captionconvergence speedImproved
Inherit VPG with randomly initialized projector (no warm-up) (ablation)OPT350M · VQAv2, GQAaccuracyHurt
Transfer from OPT125M instead of OPT1.3B to OPT2.7B (ablation)VPGTrans · COCO captionspeed-up rate: 3.3 -> 10.0Improved
VPGTrans scale-up OPT2.7B->6.7B vs original BLIP-2 OPT6.7BBLIP-2 ViT-G OPT6.7B · VQAv2accuracy: 54.3 -> 57.2Improved
VPGTrans scale-up OPT2.7B->6.7B vs original BLIP-2 OPT6.7BBLIP-2 ViT-G OPT6.7B · OKVQAaccuracy: 36.4 -> 39.8Improved
VPGTrans scale-up OPT2.7B->6.7B vs original BLIP-2 OPT6.7BBLIP-2 ViT-G OPT6.7B · GPU hoursGPU hours: 631.5 -> 59.0Improved
VPGTrans FlanT5 XL->XXL vs original BLIP-2 FlanT5XXLBLIP-2 ViT-G FlanT5XXL · VQAv2 / GQA / OKVQAaccuracy: GQA 44.7 -> 45.0; OKVQA 45.9 -> 45.0Mixed
VPGTrans FlanT5 XL->XXL costBLIP-2 ViT-G FlanT5XXL · GPU hoursGPU hours: 684.0 -> 32.4Improved
Transfer between two small LLMs (OPT350M<->FlanT5base)VPGTrans · COCO captionconvergence speedHurt

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