Reproduce “AlignGPT: Multi-modal Large Language Models with Adaptive Alignment Capability” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Fei Zhao, Taotian Pang, Chunhui Li, Zhen Wu, Junjie Guo, Shangyu Xing, Xinyu Dai
AlignGPT models varying image–text alignment and adapts it to instructions. It assigns training pairs to eight levels using CLIP similarity as a proxy, then combines a global embedding with instruction- and image-conditioned local embeddings during instruction tuning. In matched 7B comparisons, it scores 54.2 versus 50.0 on VisWiz; adaptive weighting helps on most, but not all, ablation tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.