Reproduce “Eagle 2: Building Post-Training Data Strategies from Scratch for Frontier Vision-Language Models” 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
Zhiqi Li, Guo Chen, Shilong Liu, Shihao Wang, Vibashan VS, Yishen Ji, Shiyi Lan, Hao Zhang and 19 more
NVIDIA researchers introduce Eagle2, a groundbreaking vision-language model that achieves state-of-the-art performance while using just 9B parameters, matching the capabilities of models up to 70B parameters through novel data strategies and training methods that are fully documented and shared with the research community.
Key insights
- Careful data strategy and training methods can dramatically reduce required model size
- Combining SigLIP and ConvNeXt provides superior visual processing capabilities
- Dynamic tiling enables effective handling of high-resolution inputs
- Transparent documentation of methods enables broader community advancement
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.